Skip to main content

Full text of "Introduction to descriptive linguistics"

See other formats

An Introduction to 
Descriptive Linguistics 

An Introduction to 
Descriptive Linguistics 


^OO<X>OOO<>OOOO^<>OOOOOO , ^<><>OO<>^<>$O^<X&X><£<XX>O<>0OOO 




Hartford Seminary Foundation 



Copyright, 1955, 

by Henry Holt and Company, Inc. 

Library of Congress Catalog Card Number: 55-9667 


Acc. No. . .. 494T* . 

Date.^4- . 

Call No .4.1.4 I . 


Printed in the United States of America 


Language is one of the most important and characteristic forms of 
human behavior. It has, accordingly, always had a place in the 
academic world. In recent years, however, its position has changed 
greatly: at one time the study of language was almost entirely 
restricted to specific languages, primarily those of Western Europe 
and classical antiquity; over the last few generations, a much 
broader consideration of language has taken a place at the side of 
the study of individual languages. 

As each of the social sciences has developed, it has encountered 
language problems within its domain. Psychology, sociology, and 
anthropology have each investigated language both as a type of 
human activity and as a system interacting with personality, 
society, or culture. Language has intruded even upon technological 
problems, and engineers have found themselves driven to basic re¬ 
search on human speech. Today, as a result, we have well-estab- 
lished techniques for the study of language from a number of differ¬ 
ent points of view. Each of these techniques supplements all the 
others in contributing to theoretical knowledge and the practical 
problems of the day. 

One approach has, however, received little attention until very 
recently: descriptive linguistics, the discipline which studies lan¬ 
guages in terms of their internal structures. It differs from the other 
approaches in that it focuses its attention on different facets of 
human speech. The common general subject matter and its special 
competence to handle certain types of problems bring it into impor¬ 
tant relationships with many other disciplines. 

Concurrent with the broadening of interest in language, there has 
been a fundamental change in the teaching of specific languages. 
Tongues which a past generation would have thought unworthy 
of serious attention are now taught in regularly scheduler] classes. 
The variety of linguistic structures which must be dealt with has 
increased markedly, and the need has arisen for a broader perspec¬ 
tive. Descriptive linguistics has thus become an essential con¬ 
comitant to the newer language program. 


vi Preface 

This trend is evidenced by the addition of courses in descriptive 
linguistics in many American colleges and universities. Moreover, 
the courses are having a wider influence than before. Anthropolo¬ 
gists have traditionally had some introduction to linguistic field 
methods in their training. Language majors have often had courses 
in Romance Philology or the like. These courses have been very 
different in content and outlook. But today the needs and interests 
of both groups seem to be converging, so that in many institutions 
they can meet in common courses in descriptive linguistics. 
Students in other social sciences are beginning to feel a need for a 
similar background. Linguistics courses are ceasing to be append¬ 
ages to single specialized curricula, and are attracting an ever 
more diverse enrollment. 

This textbook was written with this development in mind. It is 
not directed to prospective linguists alone; rather, widely various 
academic backgrounds and interests are assumed. Many of the 
students who use it will be particularly interested in understanding 
the place of descriptive linguistics among related disciplines, but 
will not be able to take specialized courses in these related fields. 
It has therefore been thought best to interpret the field rather 
broadly. Brief treatments of historical linguistics, dialect studies, 
communication theory, and acoustic phonetics have been included 
primarily to show their very close relationship to descriptive lin¬ 
guistics. In courses of more narrowly defined purpose or more re¬ 
stricted dimensions, these chapters can be omitted. 

This book was developed out of an introductory course at the 
Hartford Seminary Foundation, a course designed primarily as 
preparation for the language problems faced by new missionaries 
in the foreign field, but also taken by students who are starting 
preparation for specific linguistic work — analysis, translation, 
teaching, reading education, etc. The students have exceedingly 
diverse backgrounds, ranging from literature or philosophy to 
medicine or nuclear physics, and from monolingual Americans 
facing their first real language learning to fluent speakers of half 
a dozen exotic languages. The book has grown out of a syllabus, 
repeatedly revised, and more than a dozen separate experiences 
with teaching introductory linguistics. In addition, the book in a 
preliminary mimeographed version has been tried in other in¬ 
stitutions under very diverse conditions, ranging from an under- 



graduate alternative requirement to a graduate course for English 
majors. On the whole it has proven itself in a variety of situations, 
and has profited by these experiments. 

This textbook may be used in an upper-class or graduate single¬ 
semester course by omitting the more marginal chapters. With 
some supplemental reading assignments, it is adaptable for a full- 
year course in general linguistics. It should be used with the Work¬ 
book in Descriptive Linguistics, which was prepared to accompany 
this text and which gives carefully graded problems for analysis 
selected to illustrate the techniques and structures discussed and 
closely correlated with the treatment in this textbook. It is also 
desirable to have some oral instruction and practice in phonetics, 
which at the Hartford Seminary Foundation is given in drill ses¬ 
sions meeting in small groups for three hours each week. These 
drill sessions are devoted at first to drill on the English phonemic 
system and its transcription. Then attention is gradually shifted 
to sub-phonemic detail, and thence into more general phonetics. 
In the meantime, the lectures and assigned workbook problems 
are devoted to morphology. By the time Chapter 12 is reached 
and the class turns its attention to phonemics, the students have 
the minimum practical phonetic background necessary to under¬ 
stand and profit by a discussion of the phoneme principle. If the 
students have had a previous course in phonetics, a different ar¬ 
rangement might be desirable. For such students, and for instruc¬ 
tors who prefer the traditional order, Chapters 12 to 18 are written 
in such a way as to be largely independent of the morphology chap¬ 
ters. They can accordingly be assigned before Chapter 6. (Chapter 
5 might in such a case best be treated as part of the introduction.) 

A great many people have made invaluable contributions to¬ 
ward the preparation of An Introduction to Descriptive Linguistics. 
Professor J. Maurice Hohlfeld has shared with me the teaching of 
introductory linguistics for a number of years. He has read the 
drafts of each revision of the syllabus and offered innumerable 
suggestions. In addition he has assisted with much of the drudgery 
of book production. Professor W. Freeman Twadell had a large 
part in the last revision, reading the manuscript and meticulously 
criticizing the mimeographed version on the basis of his own ex¬ 
perience in teaching it. So many other linguists have helped in 
one way or another that it is almost impossible to single out a few 

viii Preface 

for mention. However, I should like to name Professors Winifred 
P. Lehmann, Raven I. McDavid, Jr., and Mark Hanna Watkins, 
who have experimented with the book in their own classes. I am 
indebted to Dr. Franklin S. Cooper and the Haskins Laboratory, 
who permitted me to use their equipment for the preparation of 
the spectrograms on page 214. Professor Ku Tun-Jou and Mr. 
Tariho Fukuda assisted with illustrations in Chapter 21. Frances 
Gleason typed and retyped the manuscript and assisted in unre- 
countable ways, not the least of which has been constant en¬ 
couragement. Finally, major acknowledgement must go to the 
students at the Hartford Seminary Foundation, who suffered 
through my mistakes and taught me, in many cases, perhaps more 
than they themselves learned. Special appreciation goes to the few 
who, with courage rare among students, told me what they thought 
was wrong with my teaching, with the course, and with the subject 
in general, and to those who encouraged me by testimonies as to 
the value of their preparation as the}' became immersed in their 
language learning. 

Hartford , Connecticut 
April, 1955 

H. A. G. 



1 Language. I 

2 English Consonants. 14 

3 The English Vowel System. 27 

4 English Stress and Intonation . 40 

5 The Morpheme. gj 

6 The Identification of Morphemes. 55 

7 Classing Allomorphs into Morphemes. 7 g 

8 Outline of English Morphology. 

9 Some Types of Inflection. 211 

10 Syntax. . 

11 Some Inflectional Categories. 143 

12 The Phoneme.jgg 

13 Phonemic Analysis.. 

14 Articulatory Phonetics.287 

15 Acoustic Phonetics. 205 

1G Interpretations of English Phonemics.221 

17 Phonemic Systems. 238 

18 Phonemic Problems in Language Learning.251 

19 The Process of Communication.266 

20 Variation in Speech.284 

21 Writing Systems.. 

22 Written Languages. 3 jg 

23 Language Classification. 333 

24 Some Languages and Language Families.350 

Selected Bibliography. 373 

Notes . 379 

Intlex .381 






1.1 As you listen to an unfamiliar language you get the impression 
of a torrent of disorganized noises carrying no sense whatever. To 
the native speaker it is quite otherwise. He pays little attention to 
the sounds, but concerns himself instead with some situation which 
lies behind the act of speech and is, for him, somehow reflected in 
it. Both you and he have failed to grasp the nature of the phe¬ 
nomenon. Neither the casual observer nor the usual native speaker 
can give any real information about a language. To be sure, some 
people, Americans perhaps more than most others, have decided 
notions about language. But the ideas held and discussed come 
far short of giving a complete picture of the language and some¬ 
times have very little relationship to the facts. Even people with 
considerable education are often wholly unable to answer certain 
quite simple questions about their language. For most people 
language is primarily a tool to be used, rather than a subject for 
close and critical attention. 

It is probably well that it is so. Yet there are important human 
problems into which language enters intimately and on which it 
exerts such a profound influence that an understanding of its 
mechanism would contribute materially to their solutions. More¬ 
over, every phase of human activity is worthy of study. Thus, for 
practical reasons, as well as to satisfy man’s innate curiosity, 
language deserves careful and intelligent study. 


2 Language 

1.2 Language has so many interrelationships with various as¬ 
pects of human life that it can be studied from numerous points of 
view. All are valid and useful, as well as interesting in themselves. 
Linguistics is the science which attempts to understand language 
from the point of view of its internal structure. It is not, of course, 
isolated and wholly autonomous, but it does have a clearly and 
sharply delimited field of inquiry, and has developed its own highly 
effective and quite characteristic method. It must draw upon such 
sciences as physical acoustics, communications theory, human 
physiology, psychology, and anthropology for certain basic con¬ 
cepts and necessary data. In return, linguistics makes its own 
essential contributions to these disciplines. But however closely it 
may be related to other sciences, it is clearly separate by reason of 
its own primary concern with the structure of language. 

1.3 What then is this structure? Language operates with two 
kinds of material. One of these is sound. Almost any sort of noise 
that the human vocal apparatus can produce is used in some way 
in some language. The other is ideas, social situations, meanings — 
English lacks any really acceptable term to cover the whole 
range — the facts or fantasies about man’s existence, the things 
man reacts to and tries to convey to his fellows. These two, insofar 
as they concern linguists, may conveniently be labeled expression 
and content. 

The foreigner who hears merely a jumble of sounds has not 
really heard the language, not even the part of it which we have 
called expression. All that he has heard is sounds, the material 
which language uses to carry its message. This is not the domain 
of the linguist, but that of the physicist. The latter can analyze the 
stream of speech as sound and learn many things about it. Ilis 
findings have both theoretical and practical importance; the de¬ 
signs of telephones, radios, and much other electronic equipment 
depends in an essential way upon such findings. They also con¬ 
tribute basic data to linguistics, and to numerous other sciences, 
including psychology and physiology, as well as to physics itself. 

The linguist is concerned with sound as the medium by which 
information is conveyed. To serve in this way, speech must be 
something quite different from the jumble of sound apparent to 
the foreigner. It is, in fact, an organized system or structure, and 
it is this structure that lies within the subject field of linguistics. 

Language 3 

The linguist analyzes speech as an orderly sequcncy of specific 
kinds of sounds and of sequences of sounds. It is orderly in terms 
of a very complex set of patterns which repeatedly recur and which 
are at least partially predictable. These patterns form the structure 
of expression, one major component of language in the sense that 
the linguist uses the term. 

The native speaker has his attention focused on something else, 
the subject of the discourse. This may be a situation which is being 
described, some ideas which are being presented, or some social 
formula which is being repeated. None of these things are language, 
any more than are the sounds which convey speech. The subject 
of the discourse stands on the opposite side and in much the same 
relationship to speech as do the sounds. The speaker comprehends 
what he is talking about in terms of an organizing structure. This 
structure causes him to select certain features for description and 
determines the ways in which he will interrelate them. It also cuts 
the situation up into portions in a characteristic way. These 
selected features, like the sounds mentioned above, also form pat¬ 
terns which recur, and which are at least partially predictable. 
These recurrent patterns are the structure of content, a second 
major component of language as the linguist treats it. 

Finally, these two structures are intimately related and inter¬ 
acting. Parts of the structure of expression are associated in 
definite ways with parts of the structure of content. The relations 
between these two complex structures are themselves quite com¬ 
plex. In every language they are different from what is found in 
every other language. The differences may be profound and ex¬ 
tensive, or they may be relatively slight. But in every instance, 
the two structures are intricate and their relationships quite 

1.4 The native speaker uses this complex apparatus easily and 
without conscious thought of the process. It seems to him simple 
and natural. But to a speaker of another of the world’s three 
thousand languages it may present quite a different picture. It may 
give an impression of being cumbersome, illogical, or even ridicu¬ 
lous. Actually, of course, the strange language is merely different. 
A true picture of language can only be had by seeing languages 
more objectively. Such a view will emphasize the immense com¬ 
plexity, the arbitrariness, and the high degree of adequacy for 

4 Language 

their purposes — features which are shared by all languages in 
spite of their divergencies. 

1.6 The dual structure of language can best be made clear by 
an example. The more technical description which will follow later 
in this book will afford more refined examples, but the following 
will indicate something of the possibilities without involving com¬ 
plicated terminology or technical concepts. 

Consider a rainbow or a spectrum from a prism. There is a con¬ 
tinuous gradation of color from one end to the other. That is, at 
any point there is only a small difference in the colors immediately 
adjacent at either side. Yet an American describing it will list the 
hues as red, orange, yellow, green, blue, purple, or something of the 
kind. The continuous gradation of color which exists in nature is 
represented in language by a series of discrete categories. This is 
an instance of structuring of content. There is nothing inherent 
either in the spectrum or the human perception of it which would 
compel its division in this way. The specific method of division is 
part of the structure of English. 

By contrast, speakers of other languages classify colors in much 
different ways. In the accompanying diagram, a rough indica¬ 
tion is given of the way in which the spectral colors are divided by 
speakers of English, Shona (a language of Rhodesia), and Bassa 
(a language of Liberia). 








red | 


cips w uka 







The Shona speaker divides the spectrum into three major por¬ 
tions. Cips w uka occurs twice, but only because the red and purple 
ends, which he classifies as similar, are separated in the diagram. 

Language 5* 

Interestingly enough, citema also includes black, and cicena white.. 
In addition to these three terms, there are, of course, a large 
number of terms for more specific colors. These terms are com¬ 
parable to English crimson, scarlet, vermilion, which are all varieties 
of red. The convention of dividing the spectrum into three parts 
instead of into six does not indicate any difference in visual ability 
to perceive colors, but only a difference in the way they are 
classified or structured by the language. 

The Bassa speaker divides the spectrum in a radically different 
way: into only two major categories. In Bassa there are numerous 
terms for specific colors, but only these two for general classes of 
colors. It is easy for an American to conclude that the English 
division into six major colors is superior. For some purposes it 
probably is. But for others it may present real difficulties. Bota¬ 
nists have discovered that it does not allow sufficient generalization 
for discussion of flower colors. Yellows, oranges, and many reds 
are found to constitute one series. Blues, purples, and purplish reds 
constitute another. These two exhibit fundamental differences that 
must be treated as basic to any botanical description. In order to 
state the facts succinctly it has been necessary to coin two new and 
more general color terms, xanthic and cyanic, for these two groups. 
A Bassa-speaking botanist would be under no such necessity. He 
would find ziza and hui quite adequate for the purpose, since they 
happen to divide the spectrum in approximately the way necessary 
for this purpose. 

1.6 Now for a simple statement of structure in the expression 
part of language: The sounds used by English are grouped into 
consonants and vowels (and some other categories). These are 
organized into syllables in a quite definite and systematic way. 
Each syllable must have one and only one vowel sound. It may 
have one or more consonants before the vowel, and one or more 
after the vowel. There arc quite intricate restrictions on the 
sequences that may occur. Of all the mathematically possible com¬ 
binations of English sounds, only a small portion are admitted as 
complying with the patterns of English structure. Not all of these 
are actually used, though the unused ones stand ready in case they 
should ever be needed. Perhaps some day a word like ring may 
appear in response to a new need. Shmoo was drawn out of this 
stock of unused possibilities only a few years ago. But ngvi would 

6 Language 

be most unlikely: it simply is not available as a potential English 
word, though it contains only English sounds. 

Six of these permissible sequences of sounds are somehow asso¬ 
ciated with the six portions into which English language-habits 
structure the spectrum. These are the familiar red, orange, yellow, 
green, blue, purple . This association of expression and content is 
merely conventional. There is no mason why six others could not 
be used, or why these six could not be associated with different 
parts of the spectrum. No reason, that is, except that this is the 
English-language way of doing it, and these are conventions to 
which we must adhere reasonably closely if we are to be under¬ 
stood. Sometime in the past history of the language, these con¬ 
ventions became established and have persisted with only gradual 
changes since. In their ultimate origins, all such conventions are 
the results of more or less accidental choices. It is largely fortuitous 
that the spectrum came to be so divided, that the specific words 
were attached to the colors so distinguished, or, indeed, that the 
sounds from which they were formed were so organized that these 
words were possible. These irrational facts, with many others like 
them, constitute the English language. Each language is a similarly 
arbitrary system. 

1.7 The three major components of language, as far as language 
lies within the scope of linguistics, are the structure of expression, 
the structure of content, and vocabulary. The latter comprises all 
the specific relations between expression and content — in the 
familiar terminology, words and their meanings. 

Vocabulary comes and goes. It is the least stable and even the 
least characteristic of the three components of language. That por¬ 
tion of the vocabulary which changes most freely is sometimes 
referred to as “slang." But even staid and dignified words are 
constantly being created and continually passing out of active use, 
to be preserved only in literature which is dated by their very 
presence. While certain types of words are more transient than 
others, none are absolutely immortal. Even the most familiar and 
commonly used words, which might be expected to be most stable, 
have a mortality rate of about twenty percent in a thousand years. 

Moreover, in the life history of an individual speaker the birth 
and death of words is very much more frequent than in the lan¬ 
guage community as a whole. Every normal person probably learns 

Language 7 

at least three words every day, over a thousand a year, and forgets 
old ones at an appreciable but lower rate. This figure must be a 
minimum, because most people have total vocabularies which 
could only be reached through even more rapid acquisition of 
vocabulary during at least part of their life. 

We have no comparable method by which the rate of change of 
content structure can be estimated. The learning of new vocabu¬ 
lary, particularly technical terms associated with the learning of 
new concepts, does of course imply certain minor changes. But it is 
quite evident that change rarely touches the most basic features 
in any given language. With regard to the structure of expression 
the facts are clearer. Few, unless they learn a second language, 
will add, subtract, or change any of their basic sound patterns after 
they reach adolescence. Grammatical constructions may increase, 
but at a rate much slower than the increase of vocabulary. Vo¬ 
cabulary is indeed the transient feature of language. 

1.8 In learning a second language, you will find that vocabulary 
is comparatively easy, in spite of the fact that it is vocabulary 
that students fear most. The harder part is mastering new struc¬ 
tures in both content and expression. You may have to free your¬ 
self from the bondage of thinking of everything as either singular 
or plural. Perhaps the new language will organize content into 
singular, dual, and plural (here meaning ‘three or more’). Or per¬ 
haps the new language will not give routine consideration to the 
matter. English speakers can never make a statement without 
saying something about the number of every object mentioned. 
This is compulsory, whether it is relevant or not. In Chinese, 
objects are noted as singular or plural only when the speaker judges 
the information to be relevant. The Chinese experience suggests 
that it actually seldom is, for that language operates with only 
occasional references to number. 

You will have to make similar changes in habits of thought and 
of description of situations in many other instances. You may, for 
example, have to learn to think of every action as either completed 
or incomplete, and to disregard the time of the action unless it has 
special relevance. The reorganization of thinking and perception 
may extend much deeper than such changes. In some languages, 
situations are not analyzed, as they are in English, in terms of an 
actor and an action. Instead the fundamental cleavage runs in a 

8 Language 

different direction and cannot be easily stated in English. Some of 
these divergencies between languages have been described by 
Benjamin L. Whorf in a series of papers which have been reprinted 
under the title Four Articles on Metalinguistics. Every student of 
linguistics or languages can profit from the reading of these 

You will also have to reorganize your habits of making and 
hearing sounds. You will have to discriminate between sounds that 
you have learned to consider the same. You will find that others, 
in clear contrast in English, function as one, and you will have to 
learn to respond to them as to one sound. Patterns which seem 
impossible will have to become facile, and you will have to learn 
to avoid some English patterns that seem to be second nature. 

The most difficult thing of all, however, is that these profound 
changes will have to become completely automatic. You will have 
to learn to use them without effort or conscious attention. In this 
learning process constant disciplined practice is essential. Special 
ability may be helpful, but probably much less so than is popu¬ 
larly supposed. An understanding of the basic principles of lan¬ 
guage structure — that is, the results of modem linguistic research 
— while not indispensable, can contribute in many ways. 

1.9 As we listen to a person speaking our native language we 
hear not only what is said, but also certain things about the 
speaker. If he is an acquaintance, we recognize him. If not, we 
identify him as male or female and perhaps obtain some idea of 
his age, his education, and his social background. A person’s voice 
serves at least two functions in communication. One is linguistic, 
in that it serves as the vehicle of the expression system of language. 
The other is non-linguistic, in that it carries information of a quite 
different sort about the speaker. 

This distinction it made, at least roughly, even by the un¬ 
sophisticated. If we are told to repeat exactly what another says, 
we will duplicate (provided our memory serves us adequately) 
every feature which is included in the language expression system. 
We can do that, if it is our own language, even without under¬ 
standing the content. In repeating we will make no effort to repro¬ 
duce anything beyond the linguistically pertinent features. If, 
however, we are asked to mimic another, we attempt to reproduce 
not only the linguistic features, but every discernible characteristic. 

Language 9 

Few can mimic with any degree of success, whereas every normal 
native speaker can, perhaps with a little practice, repeat exactly 
up to the limit imposed by his memory span. 

1.10 The most basic elements in the expression system are the 
phonemes. These are the sound features which are common to all 
speakers of a given speech form and which are exactly reproduced 
in repetition. In any language, there is a definite and usually small 
number of phonemes. In English there are forty-six. These will be 
identified and described in the next three chapters. Out of this 
limited inventory of units, the whole expression system is built up. 
In many respects the phonemes are analogous to the elements of 
chemistry, ninety-odd in number, out of which all substances are 

The phoneme is one of those basic concepts, such as may be 
found in all sciences, which defy exact definition. Yet some sort of 
working characterization is necessary before we go on. The fol¬ 
lowing is hardly adequate beyond a first introduction to the sub¬ 
ject, but will make it possible to proceed with the analysis and 
enumeration of the phonemes of English. It will be expanded and 
modified several times. Indeed, the very process of application in 
the next three chapters will constitute such emendation. 

With this in mind, we may define a phoneme as a minimum 
feature of the expression system of a spoken language by which 
one thing that may be said is distinguished from any other thing 
which might have been said. Thus, if two utterances are different 
in such a way that they suggest to the hearer different contents, it 
must be because there are differences in the expressions. The 
difference may be small or extensive. The smallest difference 
which can differentiate utterances with different contents is a 
difference of a single phoneme. This description is best illustrated 
by a full-scale application in the presentation of the phonemic 
system of a language. Since this cannot be done in brief compass, 
no illustration will be given until the English phonemes are pre¬ 
sented in Chapters 2, 3, and 4. 

1.11 There are two things about phonemes that must be ex¬ 
plicitly pointed out in anticipation of any such presentation: 

Phonemes are part of the system of one specific language. The 
phonemes of different languages are different, frequently incom¬ 
mensurable. It is for this reason that a foreigner hears only a 

10 Language 

jumble which he cannot repeat. The sounds of the unfamiliar 
language do not fit into his phonemic system, and so he can com¬ 
prehend no order in a simple utterance. If anything which is said 
about the phonemes of one language happens to apply to those 
of another, we must regard it as fortuitous. 

Phonemes are features of the spoken language. Written language 
has its own basic unit, the grapheme. Something will be said about 
this later. If, of necessity, written words are cited as illustrations, 
it must be constantly borne in mind that the written form is not, 
and cannot be, an illustration of a phoneme. Instead, it is the 
spoken form which the written form is expected to elicit which 
illustrates the phoneme under discussion. This inevitably intro¬ 
duces a major difficulty into the presentation. The illustrative 
words have been selected with the intention that they should be 
as generally as possible pronounced by all Americans in the same 
way. Undoubtedly this principle of selection fails in some in¬ 
stances because of dialect and individual peculiarities of the writer 
and the reader. Such instances will not vitiate the argument. For 
some Americans other examples might be needed, but examples 
can be found which will lead to the same results. 

1.12 The thinking that most Americans do about language is 
almost exclusively concerned with written English. A written 
language is, of course, a valid and important object of linguistic 
investigation. It can, however, easily mislead the unwary. Most 
of the misunderstandings which Americans have about language 
arise from a failure to keep clearly in mind the nature and limita¬ 
tions of a written language. 

A written language is typically a reflection, independent in only 
limited ways, of spoken language. As a picture of actual speech, 
it is inevitably imperfect and incomplete. To understand the struc¬ 
ture of a written language one must constantly resort either to 
comparison with the spoken language or to conjecture. Unfor¬ 
tunately, recourse has been too largely to the latter. Moreover, 
conjecture has been based not so much upon an intimate knowledge 
of the ways of languages in general (the results of descriptive 
linguistics) as to a priori considerations of supposed logic, to 
metaphysics, and to simple prejudice. While logic and metaphysics 
are important disciplines and can make significant contributions to 
an understanding of language, the customary manner of applying 

Language 11 

them has redounded neither to their credit nor to the elucidation 
of language structure. Linguistics must start with thorough in¬ 
vestigation of spoken language before it proceeds to study written 
language. This is true of languages with long histories of written 
literature, such as English, no less than those of isolated tribes 
which have never known of the possibility of writing. 

1.13 The second basic unit in the expression system is the 
morpheme. This again cannot be exactly defined, and several 
chapters will be given to discussing it. For the present, however, 
let us characterize a morpheme as follows: It is the unit on the 
expression side of language which enters into relationship with the 
content side. A morpheme is typically composed of one to several 
phonemes. The morpheme differs fundamentally from the pho¬ 
neme, which has no such relationship with content. That is, 
phonemes have no meanings; morphemes have meanings. 

The simpler words of English are morphemes. Other words con¬ 
sist of two or more morphemes. Like the phonemes, the morphemes 
enter into combinations in accordance with definite and intricate 
patterns. The expression structure is merely the sum of the pat¬ 
terns of arrangement of these two basic units. 

1.14 Using the phoneme and the morpheme as their basic units, 
linguists have been able to build a comprehensive theory of the 
expression side of language, and to make detailed and compre¬ 
hensive statements about the expression systems of specific 1 lan¬ 
guages. This is what is ordinarily called descriptive linguistics. 
It is the basic branch of linguistic science. Others are historical 
linguistics, dealing with the changes of languages in time, and 
comparative linguistics, dealing with the relationships between 
languages of common origin. Descriptive linguistics is conven¬ 
tionally divided into two parts. Phonology deals with the phonemes 
and sequences of phonemes. Grammar deals with the morphemes 
and their combinations. 

In some respects linguistics has developed more precise and 
rigorous methods and attained more definitive results than any 
other science dealing with human behavior. Linguists have been 
favored with the most obviously structured material with which to 
work, so this attainment is by no means due to any scientific 
superiority of linguists over other social scientists. It is also the 
direct result of the discovery of the phoneme, a discovery which 

12 Language 

allows the data to be described in terras of a small set of discrete 
units. Within a given language, a given sound is either a certain 
phoneme or it is not; there can be no intergradation. This fact 
eliminates from linguistics a large measure of the vagueness and 
lack of precision characteristic of most studies of human behavior. 
It would be presumptuous to claim that this advantage has been 
thoroughly exploited by linguists, but it is certainly fair to say 
that in some places, linguistics has achieved an appreciable measure 
of scientific rigor and has the foundations for further development 
in this regard. 

The chief evidence for the high order of development of lin¬ 
guistics as a science lies in the reproducibility of its results. If two 
linguists work independently on the same language, they will 
come out with very similar statements. There may be differences. 
Some of these differences will be predictable. Very seldom will any 
of the differences be deep-seated. Usually it will be quite possible 
to harmonize the two statements and show that by simple restate¬ 
ments one result can be converted into the other. That is, the two 
results will have differed largely in inconsequential ways, often 
only in external form. 

1.16 The content side of linguistics has developed much less 
rapidly and to a very much less impressive extent than the study 
of expression. Indeed, it cannot as yet justifiably be called a science. 
Undoubtedly this has been a source of frustration in linguistics as 
a whole. One of the greatest shortcomings of descriptive work with 
the expression aspect of language has been a lack of understanding 
of the relationships between expression and content, and the 
inability to use the analysis of content in attacking related prob¬ 
lems in expression. Here is the great frontier in linguistic knowledge 
on which we may look for progress in the next decades. 

There have been three reasons for this neglect of the content 
side. First, linguists have been late in comprehending the real 
significance of the two-sided nature of language. Their attention 
has been diverted from this basic problem by the great advances 
being made within the analysis of expression. 

Second, there has been no way to gain access to the content 
structure except through the expression structure. This requires 
an inferential method which has not appealed to linguists busy 
with building a highly rigorous method for the handling of more 

Language 13 

directly observed data. Content has therefore had an inferior status 
in the eyes of linguists. 

Third, the content, apart from its structure, has not been 
amenable to any unified study. The substance of content is, of 
course, the whole of human experience. Thousands of scientists 
have labored, each in some one of numerous disciplines, in eluci¬ 
dating this mass of material. But there is no one approach which 
can comprehend the whole and so serve as a starting point for 
comparison of the different structures which can be imposed upon 
it. Only isolated portions of the content system can as yet be 
studied as structure imposed on a measurable continuum of ex¬ 
perience. The examples of structuring of color concepts discussed 
above suggest the possibilities and make the lack of further oppor¬ 
tunities for comparison the more tantalizing. 

1.16 In contrast, the expression plane starts with much simpler - 
materials. The sounds producible by the human voice can be 
studied comprehensively by several approaches. Two of these have 
reached the degree of precision which makes them useful to 
linguistics: articulatory phonetics, a branch of human physiology, 
and acoustic phonetics, a branch of physics. These are discussed 
in Chapters 14 and 15. It is hard to imagine the scientific study of 
the expression aspect of speech attaining anywhere near the present 
degree of development without the aid of phonetics. The structure 
can be systematically described only because the underlying sounds 
can be accurately described and measured. 

The study of content structure must proceed, at present, without 
equivalent source of order in the totality of its primary data. 
Because of this it is relatively poorly developed. It is equally as 
important as the expression plane, yet it will be necessary to give 
it much less attention in this book. What little can be said is often 
semi-scientific at best. We do not even have a clear idea of the 
basic unit or units, and hence no basis for the high degree of pre¬ 
cision which characterizes the study of language expression. 




English Consonants 

2.1 The first step in studying any spoken language is to determine 
the phonemes. For a linguist studying his own native speech, this 
is relatively easy. He is accustomed to using the phonemic system, 
so his feeling about the sounds will often assist materially. For a 
person of a different background, the problem is much different. 
Some points which will seem obvious to a native speaker will 
require proof to the non-native. It is profitable in any case to seek 
rigorous demonstration, as there are occasional features which 
contradict popular prejudices. Not infrequently the difficulties 
arise from confusion of spelling and pronunciation. 

At this stage something less than full rigor will be demanded. 
The demonstration which follows is designed primarily for edu¬ 
cated Americans. A person of a different mother tongue might 
consider some points in it rather unconvincing. All the results 
which are arrived at here can be demonstrated more objectively 
by techniques which minimize the effects of the investigator’s 
language background. The basic principles of these methods will 
be presented later in Chapters 12 and 13. They will be much more 
meaningful if the student has first acquainted himself with the 
phonemic system of one language and worked with it as a tool in 
language study. 

2.2 In this chapter attention will be restricted to the conso¬ 
nants. The familiar meaning of the term will be adequate for the 


English Consonants 15 

present. That is, we are considering the sounds that are not 
ordinarily thought of as vowels. For convenience of presentation, 
we will restrict our attention in this chapter and the next to words 
of one syllable. It will also be necessary to base the discussion on 
some one form of spoken English. This will be the author’s ordinary 
pronunciation of words in his active vocabulary. Another person’s 
speech will yield identical or nearly identical results, though some 
of the particular examples used may be pronounced quite differ¬ 

2.3 To find the phonemes we must compare samples of spoken 
English that are distinct both in expression and content. We must 
be careful to ensure that both types of differences are present. 
BUI (a man’s name) and bill (a request for payment) are obviously 
different in content. But they are not recognizably different in 
expression. If one of the two is said without context, no other 
person can distinguish which one of the two has been said. The 
difference in content is not matched by a difference in expression. 
Such a pair is of no use in determining the phonemes of English. 

If two different persons say bill, the two utterances may be 
recognizably different. But the difference is not in the linguistic 
expression. One pronunciation will be considered, by a native 
speaker, as a repetition of the other. The difference between the 
two utterances conveys information about the speakers, not about 
the message. Linguistically the two are identical, and hence do not 
afford a contrast useful for our purpose. 

Occasionally it will be necessary to ascertain by experiment that 
two words are actually different. Two similarly pronounced words 
may occur in such different contexts that a native speaker may 
never have compared their pronunciation closely. To be certain, 
you can prepare a list of random repetitions of the two words in 
question. If they are different in expression, a second speaker of the 
same dialect should be able to identify them when read. If the 
two words are not distinguishable, mere chance will produce about 
one half correct identifications. 

2.4 If the two samples are bill and pill, there is both a difference 
of content and a difference of expression. The latter enables the 
hearer to recognize the word and associate it with the proper 
content. Irrespective of the context or lack of it, an American can 
distinguish these two words as they are said by any native speaker. 

16 English Consonants 

The two must therefore differ in at least one significant feature in 
the expression system, that is, in at least one phoneme. This fol¬ 
lows from the definition given in the last chapter. The phoneme 
is the minimum feature of the expression system of a spoken lan¬ 
guage by which one thing that may be said is distinguished from 
any other thing which might have been said. We will find that bill 
and pill differ in only one phoneme. They are therefore a minimal 

2.6 In calling bill and pill a minimal pair we assume that they 
differ by only one phoneme. Can we justify this? If these two words 
differed by two phonemes it might be possible to find a word that 
differs from both bill and pill but which differs from each of these in 
a smaller degree than these two differ from each other. This would 
be a word which shares with bill one of the phonemes by which bill 
differs from pill and with pill one of those by which pill differs from 
bill. Diligent search through the vocabulary of English will fail 
to reveal any such word. This, of course, proves nothing, for we 
know that only a small portion of the possible words are actually 
used. The word which we are seeking could easily be one of the 
unused possible forms. But if we could find no such word to com¬ 
pare with bill and pill, we might expect that one could be found 
to compare with some other pair which differs in a similar way. 

We can conduct a similar search for a word differing from bat 
and pat in some smaller feature than the difference between these 
two. And so likewise for numerous other pairs: tab and tap, but 
and putt, etc. No matter how many such pairs we examine, no 
evidence can be found that any smaller difference than that 
between bill and pill exists in English. This will establish that 
bill : pill is a minimal pair. 

2.6 To make a complete inventory of the consonant phonemes 
of English we will need to find a large number of such minimal 
pairs. Since we may expect that the number of phonemes will be 
considerable, this may be quite laborious. Time and effort can be 
conserved by finding sets of words which seem to be minimally 
different rather than mere pairs. One such set is the following: 


















English Consonants 1 7 

We are not concerned that many of these words show similar 
spellings, nor that some do not. Our interest is in spoken English, 
hence we are looking for sets of words which are alike in pronun¬ 
ciation. In some cases, the words of such a set will rime. 

2.7 The set of seventeen words just cited proves that the initial 
consonant in each is phoncmically distinct from each of the other 
sixteen. There is no evidence that this is the total list of English 
consonants. We may proceed by adducing additional sets of the 
same sort. These should be matched with the sets already found, 
as follows: 
































Fourteen pairs of words match in the two sets. Each set includes 
some word or words without counterparts in the other. We must 
find some evidence that the initial of yet is phonemically distinct 
from those of till, kill, fill, and rill. Perhaps the next set of words 
we find will provide this, but if not, pairs can be sought for each 
contrast. Ten : yen, tot : yacht, or two : you, etc., will prove one 
of these pairs to be phonemically distinct. Kelp : yelp, coo : you, 
call : yawl, etc.; fell : yell, fail : Yale, fen : yen, etc.; and rung : 
young, rue : you, roar : your, etc., will establish the others. 

Proceeding in this way we will gradually build up a list of 
phonemes. Before long the discovery of new additions to the list 
will become less frequent and finally cease. There are exactly 
twenty-four in most dialects of English. Search for more than this 
will prove fruitless. 

2.8 It is desirable to have some less awkward means of re¬ 
ferring to these phonemes than by saying “the initial of bill. ” We 
may provide a convenient notation by selecting twenty-four 
symbols, each of which will, by definition, refer to one and only 
one of the consonant phonemes. The most satisfactory will be the 
ordinary letters of the alphabet, eked out by a few additional signs 
of the same type. But if this system is used, symbols for phonemes 
may be confused with letters in various other uses. It will be 

18 English Consonants 

necessary to mark them as representations of phonemes. This we 
will do, following well established convention, by enclosing pho¬ 
neme symbols in / /. Thus /p/ means the phoneme found, among 
other places, at the beginning of the word pill as usually pro¬ 
nounced. Phonemes always refer to sounds, never to spellings. 
To indicate the letter we will write p. 

2.9 The twenty-four symbols selected are the following. There 
is reasonable agreement among linguists on all of these, and 
unanimity on most. Throughout the remainder of our discussion 
of English structure a knowledge of these symbols and the values 
given will be assumed. 




































































Two of these, /ij/ and /2/, never occur initially. As we have 
defined it, /h/ never occurs finally; and /w/ and /y/ do not 
occur finally in the material under consideration here. These facts 
account for the blank spaces in the table. 

2.10 It may be calculated that twenty-four phonemes will 
form 27G pairs. To prove each of the twenty-four phonemes to be 
distinct by the method discussed above, it will be necessary to 
find at least one minimal pair of words for each of the 276 pairs 
of phonemes. In most cases this is easy. For some, hundreds of 
pairs can be found readily. But for other pairs of consonants, 
suitable pairs of words are few, and many involve comparatively 
rare words. It may therefore take considerable searching to find 
them. For example, only five minimal pairs are known to me for 
the contrast /0/ : /?>/. These are thigh : thy, ether : either, mouth 
(noun) : mouth (verb), wreath : wreathe, and thistle : this’ll (the last 

English Consonants 19 

is a contrast between a word and a phrase). For /§/ : /2/, minimal 
pairs are even rarer, and only the following are known in my 
speech: dilution : delusion, glacier : glazier, and Aleutian : allusion. 
Some speakers pronounce azure to form a minimal pair with Asher. 
There may of course be other pairs which I have not yet found, 
but there certainly are not many of them. Diligent and systematic 
search has been made for both these pairs of consonants. 

At the moment of writing, there are twelve pairs of consonants 
for which I know no minimal pairs in the material we are con¬ 
sidering. These are: 

/£/ : /v § 6 ] q y w h/ 

/*/:/** ywh/ 

That these should not be found is understandable enough, though 
that does not help to prove the phonemic distinctness of the pairs 
concerned. In English, /2/ is a rare phoneme, and particularly so 
in monosyllables. The author knows only three such words, loge, 
beige, and rouge. The odds are against finding contrasts with only 
three words with which to work. /S/ also occurs in comparatively 
few words. Both /£/ and / 13 / occur in monosyllables only after 
vowels. In the material under consideration, /y w h/ occur only 
initially. This, of course, renders minimal pairs impossible. We 
will, however, later find that the distribution of /y/ and /w/ is 
much wider and that minimal pairs can be found. These are 
rather numerous for /y/ : /q/ but somewhat rarer for /w/ : /ij/. 

As mentioned above, some pairs are known for /§/ : /£/• These 
involve three-syllable words. Presumably by diligent search 
through the total vocabulary, minimal pairs might be found 
for all English consonant phonemes. But there is no guarantee 
that all will be found, and in any case, it is hardly a feasible pro¬ 
cedure. Reliance on minimal pairs would be particularly difficult 
for a foreigner whose English vocabulary might be small. Other 
techniques are available which arc more efficient and leave less 
to chance. 

2.11 The expression system of a language is a structure im¬ 
posed by the language on the sounds which are used in speech. 
The phonemes are the units in this system. They cannot be con¬ 
sidered as pure abstractions remote from the physiological and 
acoustic realities of the sounds concerned. Any further examina- 

20 English Consonants 

tion of English consonants must be based on an understanding 
of the phonetic nature of the sounds themselves. 

There are two points of view from which sounds can be profit¬ 
ably studied and described. The oldest and best established of 
these is known as articulatory phonetics. It is based on the as¬ 
sumption that the characteristics of speech sounds are the results 
of their modes of formation. They may accordingly be described 
and classified by stating the position and action of the various 
speech organs. This physiological approach has been one of the 
basic tools of linguistics for many years and has proved very 

Only recently has it been possible to study the linguistically 
significant features of speech sounds from an acoustic point of 
view. The method is so new that the full impact on linguistics has 
not yet been felt. Acoustic phonetics, however, promises to be of 
great significance as linguists gain access to the expensive equip¬ 
ment necessary and familiarize themselves with the techniques 
and with the interpretation of the results. To date, the chief 
effect has been to validate and reinterpret slightly some of the 
older ideas derived from articulatory phonetics. The following dis¬ 
cussion is necessarily based on the older methods. 

2.12 The sounds used in English arc produced by a few basic 
mechanisms combined in various ways. One of the more important 
of these is voice, a regular periodic vibration generated through 
the action of the vocal cords. The latter are two bands of elastic 
tissue in the larynx. They may be opened to permit free breathing, 
or brought together to produce various types of sounds, the most 
important of which is voice. In this instance, they are closed, 
but somewhat lightly. The air behind is moderately compressed 
by contraction of the thoracic cavity. This air pressure forces the 
cords apart. A small amount of the air passes through, and they 
again snap closed because of their elastic nature. In this way an 
alternating opening and closing of the cords is produced. This 
generates the sound waves which we call voice. Voice has a definite 
pitch or frequency, which depends on the number of openings of 
the cords per second. This frequency is controlled, among other 
ways, by tension on the cords. In singing, the control is careful and 
obvious, and with a good singer highly precise. 

2.13 The shapes of the passages between the vocal cords and 

English Consonants 21 

the outside air modify the quality of the voice in various ways. 
Sounds that involve no other mechanism than these two are called 
resonants. All English vowels and /mnqlryw/ are resonants. 
They may be divided into three groups. 

In /m n q/ the passage through the mouth is completely closed 
at some point, but the passage through the nose is open. They are 
called nasal resonants or merely nasals. The three differ only in 
the point at which the mouth passage is closed. 

In /l/ the mouth is closed at the mid-line by the contact of the 
tip of the tongue against the gums. There is an opening at one or 
both sides. It is called a lateral resonant, or merely a lateral. 

In /r y w/ the passage through the mouth is open at the mid¬ 
line. They are therefore called median resonants. This term in¬ 
cludes also all English vowels. Because of the close similarity of 
/ryw/ to the vowels, they are sometimes called semivowels. A 
more detailed description will be given in connection with that of 
the vowels in the next chapter. 

2.14 If the vocal cords are brought together sufficiently to 
obstruct the passage of air, but not to produce voice, a different 
sort of sound results. This is glottal friction. Friction differs from 
voice in that there is no clearly marked fundamental pitch. The 
effect is therefore not musical. It may be modified by the shapes of 
the passages above in the same way as voice, so that different 
whispered sounds are distinguishable, though not as clearly as 
when spoken. 

The phoneme /h/ in normal speech consists largely of glottal 
friction. Since the mouth normally is moving toward the position 
for the following vowel, /h/ is frequently similar to a whispering 
of that vowel. This means that /h/ varies widely in quality with 
its environment. It may even be a whispered nasal, as in /mhm/ 

2.15 Friction can be produced at various other points in the 
air passage by any very narrow opening. The sounds /f 0 s §/ are 
largely such friction produced at a narrow constriction in the 
mouth. They are therefore called fricatives. The sounds /v $ z 1/ 
are similar except that there is simultaneous voice. They are there¬ 
fore called voiced fricatives, and in distinction /f 0 s §/ are called 
voiceless fricatives. 

Fricatives differ also in the shape of the narrow opening in which 

22 English Consonants 

they arc produced. In /f v 0 5/ it is relatively wide from side to 
side but very narrow from top to bottom. Because of the slit-like 
shape of the opening, these sounds are called slit fricatives. In 
contrast in /s z § 2/ the opening is much narrower from side to 
side and deeper from top to bottom. These sounds are called 
grooved fricatives. 

2.16 The sounds /p t k b d g/ are formed by the complete 
closure of the air passages. There may be either closure and open¬ 
ing, or only closure, or only opening. These differences are not 
significant in English. The sounds /b d g/ are further characterized 
by voice at the moment of closing or opening; they are called 
voiced stops, /p t k/ have no voice at the moment of opening or 
closing. They are therefore called voiceless stops. English voiceless 
stops usually have a rather strong release of breath between the 
opening and the beginning of voicing for the following vowel. This 
is aspiration, and the stops are said to be aspirate. The amount of 
aspiration in /p t k/ is rather variable in most English dialects, 
but the variation is never significant. Aspiration may be nearly or 
entirely lacking in certain environments. The commonest of these 
is after /s/ in such words as spill, which should be contrasted with 
pill and bill. Aspiration may most easily be detected by holding a 
thin strip of paper before the lips. It should be so adjusted that it 
will jump for pill but not for bill. A similar failure to jump on the 
pronunciation of spill will indicate that the /p/ here is unaspirate. 

2.17 If it is desired to symbolize some feature of sound which 
is not phonemic, the transcription is enclosed in brackets [ ]. 
Thus we may write pill as /pil/ or as [p h il] and spill as /spil/ or 
as [sp“il]. The symbol [p h J represents an aspirated stop, whereas 
[p“] represents an unaspirated stop. Such a method of indicating 
pronunciation is called a phonetic transcription. One in which the 
phonemes are indicated is a phonemic transcription. The sig¬ 
nificance of the distinction will be made much clearer by the more 
comprehensive definition of the phoneme in Chapter 12. 

2.18 Affricates are stops in which the opening is relatively slow. 
They therefore are composed of a stop plus a movement through a 
fricative position: /6/ starts with a sound similar to /t/ and moves 
through one rather similar to /§/; /]/ starts with a sound similar 
to /d/ and moves through a sound rather similar to /2/. In some 
languages such sequences of sounds function as single phonemes, 

English Consonants 23 

in others as sequences of phonemes. In English the treatment as 
single phonemes seems most closely to accord with the evidence, 
though some linguists have advocated treating them as /tfi/ and 
/d2/- In any case the movement is relatively rapid and the 
fricative element is not very long. 

2.19 All the consonants are characterized by a closure or a 
decided narrowing at some point in the mouth. They may be 
classified by this point of articulation. In each case there arc two 
parts, known as articulators, which are brought together. These 
two serve to define the various articulations. The following are 
used in English: 








Lower Articulator 

(lower) lip 
(lower) lip 
tip of tongue 
tip of tongue 
front of tongue 
back of tongue 

the two 

Upper Articulator 
upper lip 
(upper) teeth 
(upper) teeth 
upper gums 
far front of palate 
velum (soft palate) 
vocal cords 

Bilabial stop 
/p/ or /b/ 


Labiodental fricative 
/f/ or /v/ 

Alveopalatal fricative 
/§/ or m 

Dental fricative 
/e/ or I'SI 


24 English Consonants 

The lower lip is much more important as an articulator than 
the upper. The term “lip” without specification may be assumed 
to refer to the lower lip. Similarly, “teeth” may be assumed to 
refer to the upper front teeth unless some others are specified. 

2.20 The consonants of English can be classified according to 
the accompanying chart on the basis of the types of sounds and 
the points of articulation: 





























































Semivowels voiced 




2.21 An examination of the phonemes of English as sounds 
rather than as units in a linguistic system reveals that some of them 
are actually quite variable in their pronunciation. A phoneme is a 
class of sounds so used in a given language that no two members 
of the class can ever contrast. This is the first modification of our 
original definition of a phoneme given in 1.10. Its full implications 
will not be discussed until a later chapter. However, one closely 
connected observation must be made here. All the sounds in any 
such class must be, in some measure, phonetically similar. This is 
the case, for example, with [p“] and [p h ]. Both are voiceless 
bilabial stops. It is also the case with the numerous varieties of 
/h/ which were mentioned. These are all voiceless and character- 

English Consonants 25 

ized by glottal friction. There is similar phonetic similarity in each 
of the classes of sounds that constitute English phonemes. 

2.22 If some measure of phonetic similarity is made a necessary 
condition for two sounds to be one phoneme, we can dismiss from 
further consideration some of the pairs for which minimal pairs 
were lacking. For example [q] and [t>] have nothing in common 
beyond the fact that both are voiced. The likelihood of their 
constituting a single phoneme is rather remote. However, [ 6 ] and 
[S] are much more similar, so that it is worthwhile to seek con¬ 
clusive proof that they are distinct. It is for this reason that 
diligent search has been made for minimal pairs bearing on the 
contrast of [ 6 ] and [ 8 ], whereas no such effort has been expended 
on some of the other pairs mentioned in 2.10. Equal effort might 
be expected to prove equally successful. 

2.23 However, it is still conceivable that extensive search might 
fail to uncover any minimal pairs for two closely similar sounds. 
In some languages, minimal pairs arc much more difficult to find 
than is the case in English, so much so that the analyst cannot 
afford to depend upon them. They are by no means necessary, but 
merely the most definitive evidence when they can be found. 
Other methods can, however, provide a quite reliable analysis. 

We may demonstrate one of these methods by assuming that no 
minimal pairs for [S] : [2] are known. A non-English investigator 
might not find dilution : delusion, glacier : glazier, Aleutian : allu¬ 
sion, even if these should occur in the speech of his informant. 
Or perhaps the speaker might not use some of these words at all, 
and so not actually have any minimal pair. This is not all hypo¬ 
thetical. I taught linguistics for two years before I became aware 
that these pairs existed in my own speech. Since [2] is rare in 
English monosyllables, it will l>e necessary, for this demonstra¬ 
tion, to relax our restriction and use words of two syllables. 

Numerous pairs can be found that differ in only one phoneme 
in addition to the sounds under discussion. One such is treasure 
[tre 2 or] : pressure [prefer]. Here there are two contrasts, /t/ : /p/, 
and [ 2 ] : [§]. The first is known to be phonemic from the existence 
of many minimal pairs, and hence is written /t/ : /p/. The second 
is under discussion; phonetic transcription is used to indicate that 
the phonemic status is unknown. 

The use of [2] and [§] in these words is not haphazard. If an 

26 English Consonants 

informant is asked to repeat these words several times, treasure will 
always have ['&] and pressure always £§]. There must be something 
about these words which determines which of the two occurs. There 
are only two possibilities. They are used consistently either because 
they are two different phonemes, or because of the presence of /t/ 
or /p/. We may assume as a hj'pothesis to test, that if a word has 
/t/, it will have [i~] rather than [s], and if it has /p/, it will have 
[5] rather than [&]. This will explain the two words treasure and 
pressure, but will not work elsewhere: pleasure has /p/ and (XI; 
and trashy has /t/ and [§]. Either we must abandon the hypothesis 
or introduce additional complications. As we examine additional 
data, we will be forced to conclude that to predict [£] : on 

the basis of other sounds in the words is either unworkable or 
unbelievably complex. They must be considered as separate 

What we have just done is to conclude that /s/ and /£/ are 
separate phonemes because they occur in similar environments. 
Minimal pairs afford more direct proof because they show the two 
sounds occurring in identical environments. The more nearly 
similar the words on which we base our argument, the more direct 
and conclusive it is. 

2.24 That /s/ and /£/ are two separate phonemes is no sur¬ 
prise to an American. He knows this without the marshalling of evi¬ 
dence. But it might be quite otherwise with a non-native. The way 
people hear sounds is determined by their language background. 
Americans are used to hearing /£/ and /§/ in similar environments, 
and so have learned to distinguish them. They are not accustomed 
to hearing [p~] and [p h J in similar environments, and have never 
needed to distinguish them. A native speaker learns to rely for the 
recognition of utterances on the phonemes of the language, and to 
disregard everything else as not pertinent. That is, he learns the 
phonemic system of the language. The task of an American student 
of linguistics is not one of learning a set of phonemes — he already 
knows them well — but of learning to make conscious formulations 
about the system, to attach symbols to the individual phonemes, 
and to lay a basis for understanding the relations between the 
English system and the phonemic systems of other languages. 




The English Vowel System 

3.1 The analysis of the vowels of English presents certain diffi¬ 
culties which are not encountered in the case of the consonants. 
These necessitate a presentation somewhat different from that 
which was used in the last chapter. Two of these peculiarities must 
be pointed out at the beginning. No discussion can be fruitful 
unless they are kept clearly in mind throughout. 

3.2 All American and most, if not all, British dialects have 
exactly the same total inventory of vowel phonemes. However, 
few words have the same vowels in all dialects. For example, east 
of the Allegheny Mountains, marry , merry , and Mary are generally 
all pronounced differently. West of the Alleghenies merry and Mary 
are usually alike, and in some areas all three may be alike. Numer¬ 
ous other comparable differences can be cited, not only between 
the speech of East and West, but also between smaller subdivisions 
in each. The total list of available phonemes is the same, but the 
distribution of the phonemes varies widely from dialect to dialect. 

Because of this it is impossible to select words that will illustrate 
the vowel phonemes to speakers of all dialects. An example that 
is excellent for one area of the country may be quite misleading 
for another. The phonemes must not be identified solely from the 
words which are stated to contain them. Preferably there should 
be oral presentation by an instructor who is thoroughly acquainted 
with the vowel system of English and this analysis of it. Lacking 


28 The English Vowel System 

this, dependence must be placed more on the phonetic description 
than on key words. Throughout this chapter, unless otherwise 
stated, my own pronunciation is the subject of analysis. This will 
prove quite different from that of most speakers from the Atlantic 
coast or the Southeast. 

3.3 What Americans usually think of as vowels are often not 
single phonemes but sequences of phonemes. In these instances, 
however, the two phonemes are very intimately related, much 
more so than a vowel and an accompanying consonant. This closely 
knit sequence of phonemes must sometimes be studied as a single 
unit. We will call it a syllable nucleus, since it serves as the center 
of a syllable. A syllable nucleus will be defined as a vowel, or a 
vowel and a following semivowel. 

3.4 Because of the intimate relationship between the parts of a 
S3 r llable nucleus, it will be easier first to identify whole nuclei as 
single entities, and then to divide them into their segments. 

We may obtain a partial list of syllable nuclei by the same 
method we used with the consonants. In this case we will obtain a 
set of one-syllable words in which the consonants are identical, and 
which are all distinguishable by a native speaker. It then follows 
that each word will contain a distinct syllable nucleus. The longest 
list known to me in my own speech is the following: 













A number of other possibilities must be eliminated. Beet is not 
distinguishable from beat, nor bight from bite. Another person might 
distinguish between some of these words, or fail to distinguish be¬ 
tween some others which I have listed as distinct; or he might have 
a different vocabulary. For example, relatively few Americans can 
consider bot 1 a parasite of horses ’ as an item in their vocabulary. 

Other lists can be adduced and compared as we did with the 
consonants. In this way our inventory of nuclei can gradually be 
increased. The maximum number that can be found will vary with 
the dialect under investigation from about fifteen to over thirty. 
Some will prove quite rare in the speech of a given person. I have at 
least two which occur with a frequency of considerably less than 
once in a thousand syllables. There is no need of a laborious 

The English Vowel System 29 

searching to determine the full list. For our present purpose the 
sample of the twelve established by our first set of similar words 
will be adequate for the next step in the analysis. 

3.6 If we compare the nuclei in bout and boat we may observe 
certain features in common. In both cases, during the pronuncia¬ 
tion of the nucleus the lips may be seen or felt to round noticeably. 
The speaker himself may feel that his tongue moves upward and 
backward at the same time. These similar movements of the vocal 
apparatus produce an audible similarity in the last portion of these 

A similar comparison of the nuclei of bite and bait yields com¬ 
parable results. In both, the tongue may be felt to rise and move 
somewhat forward. There is no rounding of the lips. There is an 
audible similarity in the last portion of the nuclei. 

Comparison of bout and bite demonstrate that the final portions 
of these nuclei constitute two phonemes. These words are in fact a 
minimal pair. The two consonants and the first element in the 
nuclei are the same. They differ only in the glides with which the 
nuclei end. We will transcribe the glide of bout and boat as /w/ f 
and that of bite and bait as /y/. The justification for using these 
symbols will be discussed later. 

3.6 We have just demonstrated by means of minimal pairs 
that four of our nuclei can be divided, since each half enters into a 
separate set of contrasts. With this in mind we may now examine 
the remaining nuclei of our sample, comparing each of them with 
these four. We will find one more with /y/, beat; the movement of 
the tongue is less prominent in this word than the others, but it 
is clearly discernible. We also find one more with /w/ f boot; here, 
likewise, the tongue movement is slight but observable, but the 
presence of the glide is most clearly seen in the lip rounding. This 
leaves six nuclei with neither /y/ nor /w/ glides. One of these, 
bought , presents some special problems; we will postpone the dis¬ 
cussion of it until later. The remaining five have simple vowels 
without glides. We will assign symbols for the phonemes as fol¬ 
lows: bit /bit/, bet /bet/, bat /bat/, but /bot/, and hot /bat/. The 
first four of these are pronounced the same in most dialects. 

3.7 The next problem is that of identifying these simple vowels 
with the first elements in the compound nuclei. We have already 
found that the first elements in bite and bout are the same. We may 

30 The English Vowel System 

start with this vowel for our first identification. There arc two 
ways in which to proceed. A very drawn-out pronunciation of bite 
will so prolong the vowel that we may more easily hear the quality. 
It may also, of course, introduce some small distortion, but prob¬ 
ably not enough to change the phoneme. We may also try de¬ 
liberately pronouncing fyf after each of the short vowels we have 
found. Both procedures will suggest that bite should be tran¬ 
scribed /bayt/ — that is, that the vowel is the same as that of hot. 
(Compare kite and cot if you do not have bot in your vocabulary.) 

Actually, most Americans pronounce /a/ slightly differently in 
/bayt/ and /bawt/ from the way they pronounce it in words like 
/bat/. We cannot expect exact identity of any phoneme in different 
contexts or even in different repetitions of a word. But here the 
difference is more regular than a chance variation. As we will show' 
later, both /y/ and /w/ exert a considerable and regular influence 
on any preceding vowel. When the simple vowels have all been 
identified, it will be found that the vowel of bite is nearer that of 
bot than to any other, and that it will have the same sort of rela¬ 
tionship to this vowel that we find in the case of each of the other 
vowel phonemes. 

3.8 Making allowances for similar differences in other cases, 
we may arrange the nuclei of our sample as follows: 













/bat / 

















We find no simple nucleus to match the vowel of boat or boot, 
nor any complex nuclei containing the vowels of bat and but. Of 
course, our sample has been a restricted one. As additional nuclei 
are found, some of them may be expected to fit into the vacant 
spaces in this tabulation. The search for nuclei should be conducted 
with this chart in mind. We must also search for additional simple 
vowels and additional types of diphthongs. 

3.9 Some of the missing nuclei will be found very readily by 

The English Vowel System 31 

examining further sets of monosyllabic words. Consider the 

lack /liek/ lick /iik/ look 

lake /leyk/ like /layk/ luck /lak/ 

leak /liyk/ lock /lak/ Lwfo /luwk/ 

Most of these can be readily matched by a native speaker with the 
members of the set we have been examining. The matching is in¬ 
dicated by the transcriptions. Only look seems new. It contains 
neither /y/ nor /w/ and seems to be a simple vowel. The vowel of 
look stands in the same sort of relationship to that of Luke as we 
have seen before in the case of, say, bot and bout. We may tran¬ 
scribe look as /luk/, and so fill one of the vacant places in the 
table in 3.8. 

3.10 It will be somewhat harder to find /o/. Most Americans, 
including myself, have this in very few words. The colloquial pro¬ 
nunciation of going to /gono/ is the commonest and most wide¬ 
spread. New Englanders often have this simple nucleus in such 
words as home /horn/ or whole /hoi/. Elsewhere these words would 
usually be pronounced /howm/ and /howl/. 

3.11 Careful search will also uncover some of the missing com¬ 
pound nuclei. Most of the remaining ones, however, are missing 
from certain dialects, or are of sporadic occurrence only. For ex¬ 
ample, I sometimes use /ow/ in words where most speakers of 
similar dialects would say /ow/. Thus road is pronounced either 
/rowd/ or /rowd/. This is the result of dialect mixture; in some 
dialects /ow/ is the usual pronunciation in these words and /ow/ 
is rare or sporadic. Such pronunciations are more widespread than 
is often realized. Somewhat less frequent is the use of the /ew/ in 
all or some of these words. 

The diphthong /sew/ is a similar case. House is /haws/ in most 
areas; in a few areas, however, it is regularly pronounced /haws/. 
In such dialects /aw/ may be rare. In still less frequent types of 
English, house may be /hews/ or /hows/. 

In New York City, bird is said to be pronounced “boyd.” This 
description is erroneous. The pronunciation that is actually heard 
is /boyd/, something clearly different from what the spelling 
would convey to most readers. The legend is the result of the 
inadequacy of our traditional alphabet to indicate pronunciation, 

32 The English Vowel System 

coupled with the unfamiliarity of the pronunciation to the un¬ 
critical observer from another dialect area. 

Every possible sequence of vowel and semivowel can be found 
in some American dialect. Some are quite widespread, occurring 
in all or most English speech; others are comparatively infrequent. 

3.12 Our tabulation in 3.8 shows only seven vowel phonemes. 
Actually, there are nine in English. The two not included there 
are difficult to identify in print. They are highly susceptible to 
dialect variation and even to individual differences within major 
dialects. Both occur, either as simple vowels or in compound nuclei, 
in most dialects, and one or both are usually quite frequent. 

Some dialects make a distinction between just as in a just judge 
and just as in He just came. If this is done, the first is usually 
/jost/ and the second is /jist/. Both of these pronunciations con¬ 
trast with gist /jist/ and jest /jest/. Other Americans use the same 
pronunciation, usually /jost/, for both words. Their doing so does 
not prove that they do not have the phoneme /»/. Indeed, it may 
be quite common in other words. The pronunciation /jist/ has 
long been singled out as an earmark of “poor English.” In re¬ 
sponse, many Americans have by considerable effort eradicated 
/jist/ from their speech, usually replacing it by the more acceptable 
/jost/. Some people who consider /jist/ inelegant nevertheless use 
it freely in unguarded moments. The pronunciation is certainly 
commoner than many people suppose. 

The word pretty as in pretty good commonly is said with /i/. 
It is quite variable in pronunciation: both /pritiy/ and /pirtiy/ 
are frequent; /portiy/ is not rare. In this context it is almost never 
/pritiy/, though in such a use as a pretty girl /pritiy/ is probably 
the commonest form. Both pretty and just will almost always be 
said with some other vowel in isolation. For a pronunciation which 
will illustrate /i/ they must be said in natural context. 

Children has two common pronunciations. The first syllable may 
by /6ild/ or /6ild/. In the South sister is commonly /sister/. For 
some speakers the name Willie /wiliy/ contrasts with will he 
/wiliy/ and wooly /wuliy/. All these illustrative words must be 
used with caution, however, since many speakers who use /i/ in 
any of these words will substitute /i/ when speaking carefully. 

The examples just cited do not illustrate the most characteristic 
uses of /»/, which are in unstressed syllables in longer words and in 

The English Vowel System 33 

certain words which do not normally receive stress in a sentence. 
Thus the second syllable of children may either be /rin/ or /ran/. 
Can is pronounced /keen/ when said in isolation, but in sentences 
it is more frequently /kin/. In these positions /if can be easily 
overlooked or confused with /i/ or /o/. A little practice is required 
to recognize and discriminate between these vowels in unstressed 
syllables. In most dialects it will be found that /i/ is one of the 
commonest vowels. It is particularly common in a very colloquial 
pronunciation, or in utterances containing many longer words. 

3.13 Simple /o/ is relatively common in the speech of eastern 
New England in such words as cot /kot/. Elsewhere in America 
these words are usually pronounced like /kat/. It is also the most 
usual pronunciation for “short o” in Southern British, but not in 
some other British dialects. In the dialects which say /kat/, /o/ is 
one of the rarest of syllable nuclei. It usually occurs in only a very 
few words and under somewhat unusual conditions. By the same 
token, /a/ is a very rare nucleus in some of the dialects which 
say /kot/. 

The vowel /o/ is somewhat more common in compound nuclei. 
Many Americans use this vowel in words like boy, saying /boy/. 
Others pronounce such words like /boy/, /o/ is also used as the 
vowel in nuclei like that of bought, though here too, /o/ occurs in 
other dialects. 

3.14 For some speakers cot and caught form a minimal pair for 
another type of glide. We write it as /h/. Caught is pronounced 
either as /kont/ or /koHt/, the latter contrasting in some dialects 
with /kot/. Unfortunately, over much of the continent no minimal 
pairs can be found for simple nuclei as against diphthongs. The 
following are some of the most likely examples. 

A good pair is bomb and balm. Both words vary considerably 
in pronunciation from place to place. In the West they are most 
frequently pronounced alike. Some Eastern dialects say /bam/ 
and /baHm/. Others say /bom/ and /baHm/. Some Southerners 
say /bom/ and /baHm/. Some speakers pronounce balm with an 

In many Eastern dialects both /se/ and /®h/ occur. The con¬ 
trast can frequently be heard in the pair can meaning ‘be able’ 
/kffin/ and can meaning ‘tin can’ /kiBHn/. Some speakers make an 
even greater difference and say /kten/ and /keHn/. These are the 

34 The English Vowel System 

pronunciations of these words said in isolation. In context, ‘be 
able’ is usually /kin/. Another frequent minimal pair is have 
/hffiv/ : halve /hanv/. 

Some Americans make a contrast in pronunciation between reel 
/riyl/ and real /hhI/. For some who do this the further com¬ 
parison of rill /ril/ can also be made. 

The semivowel /h/ occurs with each of the nine vowel phonemes; 
but, for most of the combinations, dependable key words are even 
harder to find than for those already cited. There is great variation 
in the use of these diphthongs from dialect to dialect. In most, 
nuclei with /h/ are commoner before /r/ and /!/ than elsewhere. 

Some dialects are popularly considered as “r-less.” These either 
do not pronounce /r/ after vowels, or else use a kind of /r/ which 
is so different from that of other speakers that it is often not recog¬ 
nized as an Jr/. In these dialects /a/ quite regularly appears in 
words that would have Jr/ in other dialects. Thus here may be 
pronounced /hir/, /hinr/, or /hiH/ in different areas. 

3.15 In summary, there are a total of nine simple vowels. Each 
of them can occur alone, with /y/, with /w/, or with /h/. This 
makes a total of thirty-six possible nuclei. Probably no single 
dialect has all of them, though some approach it closely. Every one 
of the thirty-six occurs, however, in some American dialect. 

The thirty-six nuclei can conveniently be exhibited in a chart. 
No key words have been inserted because of the variations which 
exist between dialects. The student is advised to fill in key words 
for all those which occur in his own speech. The result will be 
much more useful than any ready-made tabulation. 

The names of the vowel letters are quite confusing, and the 
addition of four more symbols makes matters worse. These are 
sometimes called: "digraph” for /»/, "barred I” for Ji/, "open 
0” for Jo/ and /SowaH/ for Jo/. For certain purposes, a number 
system has been found to be a more convenient method for oral 
identification of the nuclei. In this system the tens digits represent 
the vowels and the unit digits the semivowels. The numbers under 
this system are given with the transcriptions: 

10 /i/ _ 11 Jiy/ _ 12 /iw/ _ 13 Jm/ _ 

20 /e/ _ 21 /ey/ _ 22 Jew/ _ 23 /en/ _ 

30 /*/_ 31 /ffiy/_ 32 /aw/_ 33 /{err/_ 

The English Vowel System 35 

40 /»/ - 

50 M - 

60 /a/ _ 

70 /u/ - 

80 /o/ - 

90 /o/ __ 

41 /iy/ - 

51 /ay/ - 

61 /ay/ - 

71 /uy/ - 

81 /oy/ - 

91 /oy/ - 

42 /iw/ _ 

52 /aw/ _ 

62 /aw/ _ 

72 /uw/ _ 

82 /ow/ _ 

92 /ow/ _ 

43 /*h/ - 

53 /oh/ _ 

63 /aH/ _ 

73 /uh/ _ 

83 /oh/ _ 

93 /oh/ _ 

3.16 In view of the variations in pronunciation from dialect to 
dialect, it is particularly necessary to understand the mode of 
formation for the various vowel sounds. This will provide an 
essential check on the accuracy with which they have been iden¬ 
tified from the examples. 

An articulatory phonetic description would not be adequate by 
itself. Each of the nine vowel phonemes varies somewhat in pro¬ 
nunciation. Dialect differences include not only differences in the 
vowels or nuclei used in certain words, but also minor variations 
in the quality of the vowels themselves. Nevertheless, this varia¬ 
tion always occurs within certain limits, so that close attention to 
both phonetic quality and distribution will generally provide 
accurate identification of the vowel phonemes and compound 

3.17 Three variables in the positions of the vocal organs are 
particularly significant in the phonetic description of English 
vowels. The most important is the position of the highest part of 
the tongue. This varies in two dimensions. It may be relatively 
high, mid, or low. It may also be relatively front, central, or back. 
Note the difference in meaning between central (intermediate be¬ 
tween front and back) and mid (intermediate between high and 
low). These two variables provide a symmetrical charting of the 
English vowel phonemes: 




Front Central Back 

i i u 

e 3 o 

® a o 

The third variable which is of importance in English vowels is 
the rounding of the lips. In /u/ there is always moderate rounding. 
In /o/ there is usually somewhat less, but always enough to be 
noticeable. In /o/ the rounding is still weaker. In some pronun- 

36 The English Vowel System 

ciations it may be so slight as to be hardly noticeable, or even 
lacking. The front and central vowels are never rounded. 

3.18 /i/ is neither extremely high nor extremely front. The 
high front vowels of many other languages are both higher and 
nearer the front. For this reason /i/ is often difficult for foreigners. 
They often tend to use a higher vowel, one which impresses most 
Americans as more nearly /iy/. Frequently, however, this foreign 
vowel is a simple vowel, not a diphthong like /iy/. 

/u/ is likewise not as high nor as back as comparable vowels in 
many other languages. For this reason it also gives many foreigners 
difficulty. Some degree of lip rounding is always present. 

/*/ has the same height as /i/ and /u/ or may be higher than 
either. It may be produced by starting from /i/ and drawing the 
tongue backward toward the /u/ position. There is no lip rounding. 
The tongue position is distinctly higher than for /o/. Most students 
have their greatest difficulty in learning to distinguish /»/ and /i/. 
This may require considerable practice with such pairs as gist 
/list/ : just /jist/ and pretty /pritiy/ : /pritiy/, or with the con¬ 
trast between different pronunciations of such words as children 
/iildrin/ : /6ildron/ and sister /sistor/ : /sistor/. 

/a/ is somewhat variable in pronunciation, not only between 
speakers, but also between various contexts in the speech of a 
single person. The more fronted varieties are generally quite a bit 
lower than /as/, and seldom quite as far front. The more backed 
varieties may approach rather closely to /o/. In general, speakers 
who have a rather back /a/ have appreciable rounding in /o/. 
Those that lack rounding in /o/ have no variety of /a/ far enough 
back to be confused. The greatest difficulty in attempting to 
transcribe another dialect of English is, in the majority of in¬ 
stances, with the identification of the low vowels. Apparently most 
dialects have three phonemes in this territory, but the discrimina¬ 
tion of the three varies from dialect to dialect. To be certain, it is 
always necessary not only to observe the phonetic details but also 
to watch for the contrasts between the several phonemes. 

/o/ is sometimes confused with /a/ and also sometimes with /o/. 
In most dialects it is hard to find minimal pairs for the /o/ with 
either of these. Some dialects have /o ow oy oh/ but not /o ow oy 
oh/. To make matters worse, some dialects use /oy oh/ in the same 
words in which others use /oy oh/. As a result many Americans 

The English Vowel System 37 

are more or less accustomed to equating /oy/ with /oy/ and /oh/ 
with /oh/. If the nucleus of boy starts from a noticeably lower 
tongue position than does that of beau, the transcriptions /boy/ 
and /bow/ are probably correct. If these two seem to start from 
about the same place, then /boy/ and /bow/ are probable. Many 
people say boy with no trace of lip rounding such as can be ob¬ 
served in beau, or even in going to /gono/. In this case boy is 
probably /boy/, and /o/ and /o/ differ not only in tongue position 
but also in lip rounding. 

3.19 The glide or semivowel /w/ was defined in 3.5 as a move¬ 
ment of the tongue upward and backward with an accompanying 
increase in lip rounding. In /uw/ the lips start with moderate 
rounding, and this is noticeably increased. In /aw/ the lips are 
unrounded at the start, but then rounded somewhat, /uw/ ends 
with considerably more lip rounding than does /aw/. Indeed, the 
beginning of /uw/ may be more strongly rounded than the end 
of /aw/. The characteristic feature is not any given degree, but a 
marked increase in lip rounding during the course of the pro¬ 
nunciation of /w/. 

Both /y/ and /w/ involve a raising of the tongue. In /y/ the 
movement is up and forward, in /w/ it is up and backward. Neither 
semivowel represents movement to one definite position; rather, 
each is characterized by movement in a particular direction. In 
/ay/ the tongue does not usually rise as high as the starting point 
of /iy/; similarly, in /aw/ the tongue docs not rise as high as the 
starting point of /uw/. Moreover, /y/ represents a longer rise in 
/ay/ than in /cy/, and still less of a rise in /iy/. The rise in /aw/ 
is greater than that in /ow/, and in /uw/ it is even slighter. 

3.20 The symbols /y/ and /w/ were proposed in 2.9 for the 
initial consonants in yes and would. The description of the pro¬ 
nunciation of /y/ and /w/ just given will obviously not fit in these 
cases. Our definition of phonemic transcription requires that every 
phoneme have an unambiguous symbol, and that every symbol 
represent one phoneme. Writing /y/ in yes /yes/ and say /sey/, 
or /w/ in would /wud/ and do /duw/ implies that we have iden¬ 
tified sounds as members of a single phoneme. 

Before vowels, /y/ and /w/ also represent glides. They are in 
fact precisely opposite movements to those just described. In /yes/ 
the tongue starts a little higher and nearer the front than the 

38 The English Vowel System 

tongue position of /e/ and glides rapidly to the latter. In /sey/ the 
tongue starts at the position of /e/ and glides rapidly to a higher 
and fronter position. If a tape recording of /yes/ is played back¬ 
wards, the result is very similar to /sey/. Moreover, the /y/ before 
/[/ in ye /yiy/ starts much higher than that before /a/ in yacht 
/yat/. Before /i/ the glide is relatively short. Before /a/ it is 
much longer. In these respects the behavior of /y/ before vowels 
parallels the observed behavior of /y/ after vowels. The same 
things are true also of /w/. 

The differences between initial and final /y/, or initial and 
final /w/, must be taken as the consequences of their position 
rather than as any significant differences in their function in the 
language. Treating each pair as a single phoneme is thus quite 
justifiable. This is another instance of the kind of variation within 
phonemes which can be observed in all languages and within most 

3.21 The glide /h/ sometimes represents a movement of the 
tongue toward a more relaxed position. The highest point of the 
tongue may move toward a mid-central position, the position of 
rest, or there may be only a general relaxation of the tongue. The 
extent of either movement varies widely from speaker to speaker 
and from nucleus to nucleus. For many, particularly in /aH/, move¬ 
ment is hardly perceptible; but /aH/ is always distinctly different 
from /a/. When a movement of tongue position is not discernible, 
the difference is largely one of duration. That is, /h/ represents 
either glide or prolongation, or both. No contrast can be found 
between nuclei which end in a central glide and those which are 
merely prolonged. 

3.22 The fact that the /y/ and /w/ glides after vowels have 
counterparts before vowels naturally suggests that /h/ should also. 
The only possible candidate is /h/, as it is the only consonant which 
does not occur after vowels. There are certain tenuous similarities, 
but these are by no means as great as those found in /y/ and /w/. 
Since there is no clear evidence that /h/ has or has not a counter¬ 
part in /h/, linguists differ as to the direction in which the evi¬ 
dence points. I consider that keeping the two separate is the better 
solution, and hence use two symbols, /h/ and /h/. Many other 
linguists feel that the weight of the evidence is on the other side, 
and so use /h/ for both. On this basis the last column in the chart 

The English Vowel System 39 

of 3.15 will read: 13 /ih/, 23 /eh/, 33 /ffih/, etc. In any case, the 
two systems are closely equivalent, and either can be readily con¬ 
verted into the other. 

3.23 In many American pronunciations /r/ is a glide com¬ 
parable in a number of ways to /y/ and /w/. Most often the top 
of the tongue is turned upward, though other speakers obtain very 
similar sounds by different types of motions. Probably no single 
phoneme varies so widely from dialect to dialect as does /r/; in 
some pronunciations it is a trill or a flap, or one of a variety of 
other types of sounds. Certain American dialects pronounce /r/ 
after vowels so weakly that hearers accustomed to other dialects 
do not hear it at all. A few do not ordinarily pronounce /r/ in this 

The phoneme /r/ differs from /y/ and /w/ in that it is not as 
closely knit with a preceding vowel. That is, in most instances /r/ 
is not part of the syllable nucleus. Nevertheless, the relationship 
between /r/ and a preceding vowel is often closer than is the case 
with other consonants. 

In the pronunciation of many Americans, /or/ is phonetically a 
single /r/-like vowel. That is to say, the turning up (retroflexion) 
of the tongue tip which characterizes /r/ occurs throughout the 
/o/. This seems to be quite generally the case in the Western 
dialects. In Eastern dialects /or/ may be clearly divided into two 
parts. The first is a non-retroflexed vowel /o/ and the second a 
retroflexed glide /r/. These two pronunciations give quite different 
impressions, but seem to be phonemically equivalent. 



\ 4 

English Stress and 

4.1 In our discussion of the last two chapters we restricted our 
attention largely to the comparison of one-syllable words spoken 
in isolation. From this body of material we were able to obtain 
evidence for the consonant and vowel phonemes of English. The 
restricted nature of the data used rendered it impossible to discover 
some of the English phonemes. In this chapter the inventory will 
be completed through examination of a more varied body of 

4.2 If two-syllable words are examined, it will be found that 
they are of two kinds. One group is stressed (“accented” is a fa¬ 
miliar alternate term) on the first syllable and includes such words 
as going, spoken, phoneme. The others are stressed on the second 
syllable, among them obtain, because, above. Having observed such 
a difference, we should inquire as to whether it is phonemic. A 
search for minimal pairs is one method of investigation. A very 
large number of pairs can be found in which there are relatively 
minor differences in addition to the contrast in stress. Many of 
these are pairs of nouns and verbs, and most are spelled alike in 
the traditional orthography. For example present is [pr6zint] as a 
noun but [priz6nt] as a verb. In such instances, most English- 


English Stress and Intonation 41 

speaking people will feel that the difference in stress is more im¬ 
portant than the difference in the vowels. This impression does not 
constitute proof, but it does encourage further search for true 
minimal pairs. A few pairs can be found in which there is no 
difference in the vowels (for some speakers at least). One of these 
is permit; as a verb it is pronounced /permit/, but as a noun 
/permit/. However, since there are other pronunciations in use 
(e.g., /p5rmit/), these do not constitute a minimal pair for all 
speakers. Another is pervert, /pervert/ and /pervert/. 

These prove that stress is phonemic in English; thus two addi¬ 
tional phonemes have been established. We will call /'/ primary 
stress, and /"/ we will call weak stress. We will follow a conven¬ 
tion that /V will not be written in a transcription of words or 
longer portions unless it is desired to call special attention to it. 
In an example where stresses are shown, any vowel with no stress 
marked will be understood to have weak stress. 

4.3 Having determined that stress is phonemic, we must now 
turn back to the one-syllable words we considered in Chapters 2 
and 3 and determine whether they have stress. One way of doing 
this is to read a list of words, alternating monosyllables and disyl¬ 
lables, for example, bill, permit, pill, present, beat, etc. These must 
be read with a slight pause between them so that they will be, in 
effect, isolated. If this is done, it will be evident that the stress 
on the monosyllables will match closely that on the more heavily 
stressed syllable in the longer words. We accordingly conclude that 
all monosyllables, when said in isolation, have primary stress. This 
means that a transcription of bill as /bil/ is not complete. It should 
be /bll/. 

The stress was not discovered in our first comparisons because 
all words said in isolation have one /'/; hence no stress con¬ 
trast — minimal or otherwise — is possible in monosyllables. The 
full phonemic system of English cannot be considered as estab¬ 
lished until we have examined the full range of types of utterances 
which are amenable to linguistic study. 

4.4 The next type of data that should be examined is words of 
more than two syllables. These will easily enable us to discover a 
third degree of stress. In this case it will prove more difficult to 
discover suitable minimal pairs, but many very suggestive cases 
can be found. A technique comparable to that used in 2.23 to 

42 English Stress and Intonation 

establish the phonemic status of /§/ and /2/ will prove that this 
third degree of stress is phonemic. We will mark it /'/ and call it 
tertiary stress. It may be heard in such words as dictionary 
/dikson^riy/ or animation /amimSysin/. 

Having discovered the contrast between tertiary and weak 
stress, it is necessary to go back and re-examine the two-syllable 
words. It will be found that the more weakly stressed syllable 
most often has /“/, but not infrequently has /'/• For example 
contents is usually pronounced /k&ntfents/, very rarely /k&ntints/. 

4.5 These two cases show how essential it is to restudy all data 
in the light of each new discovery. In each of them, phonemic 
distinctions which could not easily have been discovered through 
work with limited data have proved to be easily identifiable when 
other data has suggested their existence and nature. This is a 
frequent experience in linguistics. For example, to suggest that 
stress is phonemic only in polysyllabic words will produce many 
unnecessary complications. Experience has shown that the most 
satisfactory policy is to consider that, if stress is phonemic at all, 
it is phonemic throughout English. Conversely, no feature which 
is not established as phonemic can be considered as linguistically 
significant even in a highly specialized context. The discovery of 
any new phonemic contrast imposes the obligation to examine the 
data to determine the full range of its occurrence. 

4.6 A contrast of another sort may be discovered through com¬ 
paring such sequences as night rate with nitrate. The average Amer¬ 
ican will hear the difference as a break between night and rate, 
contrasting with no break in nitrate. Since this break serves to 
distinguish utterances, it is a phoneme. It may be called open 
transition and transcribed /+/. 

Having found /+/, the principle just mentioned compels us to 
examine the language closely to find how widely it is distributed. 
It is reasonable to expect that it will be found in many more 
places than the few minimal pairs would indicate. It will prove 
convenient and helpful to search next for pairs which, while not 
minimal, show the contrast in similar environments. One such case 
is, for many speakers, minus /maynis/ : slyness /sl&y+nis/. Such 
pairs will help us to discover the characteristics by which we 
recognize / + / in different contexts. We will then find it easier to 
identify it as we begin to examine an increasing range of data. 

English Stress and Intonation 43 

By this procedure we will find that /+/ is, in fact, a frequently 
occurring feature of English pronunciation. 

The phonetic nature of / + / is complex, and many of the details 
cannot be mentioned at this time. When /+/ follows immediately 
after a syllable nucleus, as in slyness, it is expressed, in part, by a 
prolongation of the syllable nucleus. This is certainly the most 
noticeable difference in such a pair as minus : slyness. After certain 
consonants, notably /m n 13 /, it is also expressed by prolongation 
of the preceding consonant. After some other consonants it takes 
the form of a weakening of the voicing. After voiceless stops it is 
shown by contrasts in the degree of aspiration. /+/ also has certain 
noticeable effects on the following vowel or consonant. In short, 
in a pair like night rate and nitrate, though a native speaker inter¬ 
prets the facts as a break or even a very short pause between night 
and rate, the differences are actually in the details of pronunciation 
of both the /t/ and the /r/. 

4.7 Most of the traditionally defined words of an utterance are 
separated from their neighbors by a /+/, a fact which has misled 
some to consider it as more or less identical with the traditional 
word division. Nothing could be more confusing than to do so. 
Word divisions are part of our system of spelling conventions. 
Why, for example, is cannot one word, when must not is two? As a 
spelling convention, word division follows its own, largely ar¬ 
bitrary, rules. These are uniform, except for minor departures, for 
all educated Americans. Arbitrariness is the price of uniformity. 

The /+/ is of an entirely different nature. It is a feature of 
pronunciation. As such it varies from speaker to speaker, and 
indeed one speaker may be quite inconsistent in his usage at cer¬ 
tain points. / + / is a phonemic feature. As such it may be written in 
a transcription only because it is heard in the pronunciation of the 
speaker being observed. 

Actually, /+/ will be found in many places where word divisions 
customarily occur. The correlation is far from exact, however. 
Certain commonly written word divisions are almost never paral¬ 
leled by / + / in natural speech. Certain words are commonly pro¬ 
nounced with /+/ within them. One case is unknown, usually 
/dn+n 6 wn/. Some others vary: Plato is pronounced by some as 
/plfiytdw/, by others as /pl&y+tdw/. As we shall see later, the 
occurrence of /+/ is associated in definite ways with the stresses. 

44 English Stress and Intonation 

4.8 When the discussion turns from words to whole sentences 
a new difficulty arises. The necessary data cannot be presented in 
writing. A cited sentence may be read in a variety of ways. In 
actual speech a native speaker would not be in the least capricious 
in his selection of a certain intonation for a given sentence. Nor 
will the average American fail to react differently to sentences 
which are alike in the words composing them but different in 
intonation. The latter differences are correlated with differences of 
content, and hence are clearly within the domain of linguistics. 
The difficulty of presentation is merely that we have no feasible 
means of informing the reader which pronunciation is intended. 
What we call a “sentence” in written English may stand as a 
written symbolization of several different spoken utterances. The 
distinction between them cannot be indicated in such a way that 
the reader will know what spoken sentence is cited as an example. 

The result can easily be that the wrong reading is selected. The 
example then fails to prove the point at issue, or may even seem to 
contradict the description flatly, so that the reader remains un¬ 
convinced. Even experienced professional linguists have been 
grievously misled in this way and have failed utterly to understand 
presentations of facts about English intonation which are per¬ 
fectly clear when the author presents his data orally. 

It will, therefore, be necessary to use a different approach for the 
remainder of this chapter. Instead of adducing evidence from 
which the phonemes can be demonstrated, it will be better to state 
the conclusions first and then present the supporting or illustrative 
data. Your instructor will, of course, demonstrate these features 
in detail, perhaps in a manner more nearly in accord with the 
presentation up to this point than the printed text can be. 

4.9 The nature of this difficulty is shown by the very existence 
of a special art, the oral interpretation of literature, which pro¬ 
motes the ability to select the oral rendition which will be accepted 
by the native speaker as most suitable for the context. That it is 
an art, rather than a science, demonstrates that there is no re¬ 
liable method currently in use by which an author can inform his 
readers how he wants his work to be pronounced. If there were, 
expressive reading could be largely mechanical. As it is, mechanical 
reading is far from expressive. Moreover, if the public did not 
react differently to different pronunciations, there would be no 

English Stress and Intonation 45 

need to even attempt to make a selection. That is, the art would 
be unnecessary if it did not involve the manipulation of some 
features of pronunciation (phonemes) which are significant in the 
structure of the language. 

4.10 We have established the existence of three phonemic 
degrees of stress. There is one more which we will write / A / and 
call secondary. It may be heard in an ordinary pronunciation of 
such a sentence as I'm going home. Normally there will be only 
one /'/ in such a sentence. Most frequently it will be on home, but 
it may be on going. Two possible pronunciations are (marking the 
stresses on only these two words) I'm gding hdme. and I'm gding 
Mme. These have different meanings and hence constitute a 
minimal pair establishing that / A / is distinct from /'/. 

The contrast of / A / with /'/ may be seen in such a pair as 
bldek bird : bldckbird. The first implies a bird that is described 
as being black; the second a particular kind of bird which, in¬ 
cidentally, may or may not be black. Thai white bird is an albino 
bldckbird. makes sense if the stresses are as indicated, but nonsense 
if we substitute black bird. 

4.11 The four stresses /' A ' ~/ and /+/ are closely related and 
form a special system within the phonemic structure of the lan¬ 
guage. There is always at least one /+/, or some higher-ranking 
break, between every successive pair of / A / and /'/. Bldek bird 
must have a /+/ between the /'/ and the / A /. Any American other 
than a phonetic virtuoso will inevitably either pronounce a /+/ or 
change one of the stresses. /blc6k + b5rd/ or /bls&kbiird/ are both 
possible, but /*bli6kb3rd/ is not. (Forms which are impossible or 
unknown are marked with *.) It does not, however, follow 
that there must be a /'/ or / A / between every two /+/• Both 
/blifck+bard/ and /bli6k + b3rd/ are possible. Few words cited in 
isolation have both /'/ and / A /. This is because few words con¬ 
tain a /+/, and °nly such words could have both. Moreover, the 
few words that do have /+/ within them do not necessarily have 
/'/ and / A /; they may have /'/ and /'/ or various other com¬ 

4.12 Some of the breaks in a long utterance will be markedly 
different from /+/• There will be exactly one of these between 
every pair of /'/, and it will always be found at a place where a 
/ + / might otherwise occur. These breaks divide the utterance into 

46 English Stress and Intonation 

portions which are characterized by two prominent features: the 
break at the end, and the presence of a /'/• Such a unit may be 
called a clause. The term is here used to refer to a unit in the 
spoken language which is evident from the pronunciation only. It 
may frequently prove to be equivalent to a clause in the conven¬ 
tional grammatical sense, but it will often be different. The breaks 
marking the ends of clauses are clause terminals. They may 
easily be shown to be phonemic because there are three different 
terminals, and these contrast. 

Clause terminals are popularly referred to as “pauses.” There 
may be an actual pause, but this is not necessary. If there is a 
pause, it is always preceded by one of the characteristic English 
clause terminals. The clause terminal is, properly speaking, a 
means of ending a clause, not of separating two clauses, and for 
this reason one such terminal will be found at the end of the last 
clause in an utterance. 

Clause terminals are of three kinds: 

/\/ fading: a rapid trailing away of the voice into silence. Both 
the pitch and volume decrease rapidly. 

/// rising: a sudden, rapid, but short rise in the pitch. The 
volume does not trail off so noticeably, but seems to be 
comparatively sharply cut off. 

/—>/ sustained: a sustention of the pitch accompanied by pro¬ 
longation of the last syllable of the clause and some dimin¬ 
ishing of volume. 

4.13 The clause terminals are heard at the very end of the 
last syllabic in a clause. Over the remainder of the clause, the 
pitch varies in such a way that four contrasting phonemic levels 
can be recognized. The normal pitch of the voice of the speaker is 
/2/, called mid. It varies, of course, from speaker to speaker. 
Moreover, most people raise the pitch somewhat when they are 
speaking more loudly, and at various other times. Pitch /2/ is 
relatively common and serves as a standard of comparison for 
the others. Pitch /!/, called low, is somewhat lower, perhaps two 
or three notes below /2/, but the interval will vary from speaker 
to speaker and from time to time. Pitch /3/, called high, is about as 
much higher than /2/ as /2/ is above /l/. Pitch /4/, called extra 
high, is higher than /3/ by about the same amount, or may even 
be somewhat higher. /4/ is much less frequent than the other 

English Stress and Intonation 47 

three. In any utterance of any length /I 2 3/ will all be heard. 
The native speaker will identify them with little difficulty by 
comparing the three commonly occurring pitches in the utterance. 

Some variation of pitch within these four levels will be noticed 
by some observers. Others (i.e., native speakers) may have great 
difficulty in convincing themselves that this variation exists. It 
is not a significant variation, as can be demonstrated by techniques 
of the sort presented in Chapter 13. In any case, no minimal con¬ 
trasts can be found for any other pitches, whereas they can be 
found between all of the pitches here considered as phonemic. 

4.14 Unfortunately, two systems of numbering the pitch 
phonemes have arisen. A minority of American linguists use /!/ 
for extra high, /2/ for high, /3/ for mid, and /4/ for low. 

An alternative graphic system avoids the confusion of these com¬ 
peting notations and has certain pedagogic advantages as well. On 
the other hand, it has various limitations, not the least of which is 
that it is wasteful of space. It is as follows: 


extra high 

Clause terminals can be indicated by arrow heads continuing the 
line. In some of the examples below, both systems will be used: 
/I 2 3 4 \ —*/ with phonemic transcriptions, and graphic 

marking with the orthography. 

4.15 The pitch is frequently the same over long sequences of 
syllables (phonemically the same, that is: there will usually be 
sub-phonemic variation from syllable to syllable). It seems best 
to consider most such instances as containing only a single occur¬ 
rence of the pitch phoneme concerned. In transcribing spoken 
English, we need therefore only note the pitch at certain critical 
points. One of these is at the beginning of the syllable which 
has /'/. This syllable will commonly have a different pitch from 
that of the preceding syllables, usually a higher pitch. In the 
instances where it is the same, it should be noted anyway. It will 
be indicated in a phonemic transcription by a raised numeral 
immediately before the syllable. 

If the clause contains any syllables before the /'/, these will 
usually have a single continuing pitch; most often, but by no 

48 English Stress and Intonation 

means always, this will be a /2/. An indication of this pitch should 
be written at the very beginning of the clause. 

Another critical point is at the end of the clause immediately 
before the clause terminal. The pitch here will be indicated by a 
raised numeral immediately before the symbol for the terminal. 

Normally, these three are all the points at which contrasts can 
be found. It is therefore necessary to symbolize pitch at these 
places, and only these. If the /'/ is on the first syllable, there must 
always be two pitches indicated. (They may be the same, of 
course.) If the /'/ is not on the first syllable, three must be in¬ 
dicated. A few speakers have occasional patterns in which a rise 
in pitch occurs at a syllable with / A /; in these cases, four pitches 
must be marked. 

4.16 Frequently the /'/ is on the last syllable of the clause. In 
this case, the pitch frequently glides from one level to another on 
this one syllable. In most instances the pitch will fall. If the /'/ is 
farther from the end, the fall may be spread over all the intervening 
syllables, or may occur anywhere between the /'/ and the end. 
No contrasts can be found in the time, rate, or manner of this fall, 
so no details need be recorded in a phonemic transcription. 

4.17 The four pitches and three clause terminals constitute 
another sub-system within the total phonemic system of English. 
Every clause is marked and held together by an intonation contour 
consisting of two, three, or four pitch phonemes and one clause 
terminal. Intonation contours are not, of course, phonemes, but 
morphemes. Their mention in a chapter on English phonemes is 
logically out of place, but it will be somewhat easier to cite intel¬ 
ligible illustrations if this concept is introduced now. Of course, 
examples of morphemes are examples of the phonemes of which 
they are composed, but the two must be kept clearly dis¬ 

4.18 The commonest intonation contour in English is 
/(2)3IN*/. The parentheses indicate that the /2/ will occur if there 
are syllables before the /'/, but not otherwise. That is, /231\/ 
and /3I\/ are considered as variants of the same morpheme con¬ 
ditioned by the context. This contour may be used in different 
ways with what is, apart from stress and intonation, the same 
utterance. Each of the following examples has a slightly different 
meaning, as will be evident to any native speaker. 

English Stress and Intonation 49 

/^J'm + g6\viq +3 h6wm' \ / 
/*Aym + *g6\viq + h6wm^» / 
/ 3 aym + gdwiq + h6wm , '>» / 

Ihe /(2)31\/ intonation contour docs not necessarily indicate a 
statement. It is quite commonly used with questions. 

/ 2 hwfcn + oryo +3 g6 wiq+howm 1 ^/ When are you f gcnrujhome? 

Another common contour is /(2)32—►/. This usually indicates 
some close connection between the clause so marked and the 
clause closely following. 

/*&y + woz +3 g6 wi i) + h6wm*"* 5 b5t + it +3 r^yiid'^ / 

The contrast between /\/ and //•/ can be seen in a question 
such as that below. The first of these pronunciations is somewhat 
less polite or deferential than the second, and the third commonly 
means something like ‘Did you say, “What arc we having for 

/ 2 hwSt + or + wiy + h&viq+for +3 dfnor ,N »/ 
/^wH+or^riy+hjfeviQ+for^dfnor 1 ' / 
/ 2 hwdt + or + wiy + h&vij)+for^dlnor‘'y 

Both /22/"/ and /11\/ are used to mark the person addressed. 
They differ primarily in politeness. /23//, as in the third example 
below, asks a yes-or-no question repeating the question in the 
preceding clause. 

/ ! hwdt + or + wiy+h£viq + for +3 dfnor s ^mSbor 2 Sf 
/ 2 hwdt + 3r + wiy + h^viij + -for +3 dfnor J-M mjSor 1N ./ 
/ J hwdt + 8r + wiy + h^vio + for +3 din9r* ^mStin *'/ 

50 English Stress and Intonation 

How clear the contrast between these contours may be is seen 
by switching them in the following way: 

/ 2 hw^f f or + wiy + h^vii 3 + far + 3 dinor 2 ^*m45ar* / 1 
/ 2 hwdt + ar + wly + h&viQ + for +a dfnar 2 '* 2 mStin 2 '* / 

Pitch /4/ is much rarer than the others. It occurs most com¬ 
monly in utterances characterized by at least mild emphasis or 
expressing surprise. Contrast I'm going home, said as a matter-of- 
fact statement with the same words said impatiently by a person 
who has started home but is being nagged to move a bit faster. 

He might say: 

/ 2 &ym H g 6 wiij + h 6 wm | N / 

rm\ ~gotng s home. 

4.19 We may now list the complete inventory of English 
phonemes as follows: 

24 consonants /pbtdkgfiJfv 0 tSsz§imnqlrwyh/ 

9 vowels /i e a i a a u o o/ 

3 semivowels /y w h/ (/y w/ were listed also with the con¬ 


1 open transition / + / 

4 stresses /' A ' 7 
4 pitches /I 2 3 4/ 

3 clause terminals /\ / —*/ 

46 total (eliminating duplications in the listing) 




I 5 

The Morpheme 

5.1 In the last three chapters the basic elements of English 
pronunciation were shown to be forty-six phonemes. By describing 
these in detail (an undertaking not attempted here), and stating 
the characteristic distribution of each (which was merely hinted 
at), a great deal can be said about the English language. By re¬ 
cording the phonemes occurring in it, any possible English utter¬ 
ance can be so identified as to be exactly repeatable from the 
written record alone. 

These are valuable results, and an essential part of any full 
description of the language; but they fall far short of a complete 
analysis. No matter how far this line of investigation is pursued, 
nothing is revealed about the meanings of utterances in the 
language. Yet the social function of any language is to carry in¬ 
formation from speaker to hearer. Without this, speech would be 
socially useless and presumably would not exist. A phonologic 
study of language, no matter how detailed, can tell us nothing 
about meaning, because the phonemes themselves have no direct 
connection with content. They are merely the units by which the 
speaker and hearer identify the morphemes. For any further study 
of language, the morphemes and combinations of morphemes must 
be examined. When this is done, the analysis of language structure 
proceeds on a fundamentally different plane. 

6.2 Morphemes are generally short sequences of phonemes. 



52 The Morpheme 

These sequences are recurrent — but not all recurrent sequences 
are morphemes. For example, the sequence /in/ occurs thirteen 
times in a reading of the preceding paragraph; /ov/ occurs ten 
times. Such sequences as /in/ and /ov/ can profitably be studied 
as phenomena of English phonology, and some important gen¬ 
eralizations can be made about these and similar sequences. Such 
a study of /in/ about exhausts everything worth serious attention 
regarding this sequence. Not so in the case of /ov/; for /ov/, in 
addition to being a sequence of phonemes, is in each of these ten 
occurrences a morpheme, and so participates in a higher level of 
organization. On this plane /in/ is not relevant. The fact that /in/ 
is commoner than /ov/ does not affect the situation. 

5.3 The difference between /ov/ and /in/ rests on the fact that 
/ov/ in each of its ten occurrences above has a meaning — that is, 
some connection with some clement in the structure of the content 
aspect of the language — while /in/ does not have meaning, except 
as it forms a fragment of certain sequences, such as /kin/ can. 

As a morpheme, /ov/ also has demonstrable relationships with 
other morphemes in the language. These are of two kinds: In the 
phrase study of language, there are certain significant relationships 
between /ov/ and the morphemes which precede and follow it in 
this particular fragment of an utterance. These are features of this 
sample of English as such. There are also certain more general 
relationships of the morpheme of which are not limited in this way, 
and so constitute a part of the system of the language as a whole. 
These are generalizations arrived at by comparing study of language 
with many other similar sequences. For example, of can be fol¬ 
lowed by a noun, but not usually by a verb. In some constructions, 
of can be replaced by on. Compare the hat of the man with the hat on 
the man. In another way, of can be replaced by’s. Compare the 
hat of the man with the man's hat. These broader relationships are 
the subject of study in the division of linguistics known as 

5.4 The morpheme was mentioned in 1.13 as the second of the 
two basic units in linguistics. No definition was given, and it was 
stated that an exact definition is not feasible. Perhaps the best that 
can be done is to define the morpheme as the smallest unit which is 
grammatically pertinent. But it would then be necessary to define 
grammar as the study of morphemes and their combinations. This 

The Morpheme 53 

is obviously circular and hence is no definition. Nevertheless, it 
docs serve to point out something significant. As a basic concept, 
a morpheme cannot be defined beyond some such circular state¬ 
ment. In place of a definition, therefore, we must merely describe 
certain features of morphemes and give some general rules for their 
recognition. This we will do here and in succeeding chapters. 

6.6 Some morphemes can be usefully described as the smallest 
meaningful units in the structure of the language. A more precise 
statement would, of course, be in terms of relationship between 
expression and content, but for the present purpose a less exact 
statement is convenient. By “smallest meaningful unit” we mean 
a unit which cannot be divided without destroying or drastically 
altering the meaning. For example, /streynj/ as in strange is a 
morpheme; as a whole it has meaning. If it is divided, we obtain 
fragments such as /str/ or /eynj/, which have no meaning, or 
/stray/ as in stray or /streyn/ as in strain, which have meanings 
which are not significantly related to that of /streynj/. Any divi¬ 
sion of /streynj/ destroys or drastically alters the meaning. There¬ 
fore, /streynj/ qualifies under our description of a morpheme as 
the smallest meaningful unit in the structure of the language. 

However, /streynjnis/ as in strangeness is not a single morpheme, 
though it docs have meaning. It may be divided into /streynj/ 
and /nis/. Each of these pieces does have meaning, and the mean¬ 
ing of the combination is related to the meanings of the two pieces. 
Therefore /streynjnis/ is two morphemes. 

6.6 A morpheme is not identical with a syllable. The morpheme 
/streynj/ happens to be a syllable, and so are many English 
morphemes. But /konetikit/ as in Connecticut is a single morpheme, 
though it contains four syllables. Both /gow/ and /z/ in goes arc 
morphemes, though together they are but a single syllable. Mor¬ 
phemes may consist of one or several whole syllables, parts of 
syllables, or, in fact, any combination of phonemes without regard 
to their status as syllables. 

6.7 A morpheme may consist of only a single phoneme. The 
/z/ in goes just cited is a case. But the phoneme /z/ and this 
morpheme are by no means identical. The phoneme occurs many 
times where it has nothing to do with this morpheme. Instances 
are zoo /z&w/ and rose /r6wz/, both of which contain /z/ but have 
no meaning in common with the /z/ in goes. Most English mor- 

54 The Morpheme 

phcmes are intermediate in size between /z/ and /streynj/, and 
consist of two to six phonemes. 

6.8 Frequently two morphemic elements are alike in expres¬ 
sion but different in content. Such pairs are said to be homopho- 
nous, literally “sounding alike.” Thus /z/ is a morpheme both in 
goes /g 6 wz/ and in goers /g 6 worz/, but not the same morpheme, 
/z/ meaning ‘third person singular actor’ and /z/ meaning ‘plural’ 
are homophonous. Sequences of morphemes can also be homopho- 
nous, either with other sequences or with single morphemes. 
Compare /rowz/ in He rows the boat. They stood in rows, and That 
flower is a rose. 

6.9 If the morpheme is to be described as the smallest mean¬ 
ingful unit in the structure of a language, care must be taken not to 
misconstrue the words “meaningful” or “meaning.” “Meaning” 
is intended to represent the relationship which exists between 
morphemes as part of the expression system of a language and 
comparable units in the content system of the same language. A 
morpheme is the smallest unit in the expression system which can 
be correlated directly with any part of the content system. 

Using the term meaning in its ordinary familiar sense without 
careful control will in some cases be quite misleading. In many 
instances, however, it will serve as a workable approximation, if 
used with caution. For example, cal may be said to have a meaning 
since it refers, among other things, to a specific kind of animal. 
But it is also used of humans with certain personality character¬ 
istics. In a like sense, go may be said to have a similar kind of 
meaning, since it refers (among other things) to a motion of an 
object. But it is difficult, even fruitless, to attempt to specify 
exactly what motions are indicated. Compare He goes home. John 
goes with Mary, and The watch goes. Indeed, it may be used of a 
quite immobile subject as in This road goes to Weston. These varia¬ 
tions of reference to the outside world can in part be accounted for 
by the assumption that a speaker of English has learned to struc¬ 
ture content in such a way as to bring these diverse elements of 
experience together into a single category. The meaning of go rests 
in the interrelationship between the morpheme /gow/ and the 
point within the content system where these things are brought 

6.10 The content system of a language is not directly ob- 

The Morpheme 55 

servable, so that we can only with great difficulty check any such 
statement as that just made. It does, however, serve this useful 
function: It should be a distinct warning against relying on transla¬ 
tions to get access to meanings. If the structure of content imposes 
a filter between the expression system and human experience, 
translation must impose two such. Translation can only be accurate 
where the content structures of the two languages coincide. Such 
places are too infrequent to be depended upon. Where translation 
must be used (and there are many such instances in practical 
language work) the user must be constantly alert against its 

6.11 With some morphemes, meaning in the sense of reference 
to human experience outside language is wholly or largely lacking 
Consider to in / want to go. The elements 7, want, and go are 
referable, through the intermediary of English content structure 
to aspects of human experience. But it is impossible to find a 
specific factor in the situation which can be considered as the 
"meaning” of to. Nevertheless, to does have a function, since 
without it *7 want go. means nothing. (The symbol * is used to 
indicate that a form cited is either unattested or known to be 
impossible.) To merely fulfills a requirement of English structure, 
in that want cannot be followed by go without to. Such a function 
cannot be included within the traditional meaning of "meaning,” 
but in the sense in which we arc using it (the interrelationship be¬ 
tween expression and content), "meaning” — with a little stretch¬ 
ing, perhaps — can comprehend it. 

6.12 The meaning of cat might be explained (partially, to be 
sure) to a non-English speaking person by pointing out the animal 
to which it refers. It would not be possible to explain to in this way. 
Instead, it would be necessary to cite a number of cases of its use, 
and thereby point out the contexts in which it occurs regularly, 
those in which it may occur, and those in which it cannot occur 
(e.g., *7 can to go.). That is to say, to has a characteristic distribu¬ 
tion. For the foreigner, this distribution is the most easily ob¬ 
servable feature of such a morpheme, and hence the chief clue to 
its meaning. 

Morphemes like to are not alone in having a characteristic dis¬ 
tribution. Every morpheme has. Cat may occur in 7 saw the -. 

but not in 7 will - home. Go can occur in the second, but not in 

56 The Morpheme 

the first. The distribution of the morpheme is the sum of all the 
contexts in which it can occur in contrast to all those in which it 
cannot occur. A full understanding of any morpheme involves 
understanding its distribution as well as its meaning in the familiar 
sense. It is partly for this reason that a good dictionary always 
cites instances illustrative of usage. One that does not is of very 
restricted usefulness, or even very misleading. 

5.13 Morphemes can be identified only by comparing various 
samples of a language. If two or more samples can be found in 
which there is some feature of expression which all share and some 
feature of content which all hold in common, then one requirement 
is met, and these samples may be tentatively identified as a 
morpheme and its meaning. Thus boys /b5yz/, girts /gSrlz/, roatls 
/r6wdz/, etc., are all alike in containing s /z/ and meaning ‘two 
or more.’ We therefore identify s /z/ as a morpheme meaning 
‘ plural.’ This is not actually sufficient. In addition there must be 
some contrast between samples with similar meaning and content, 
some of which have the tentative morpheme and some of which 
do not. Comparison of boy /b5y/ will serve to confirm the example 
we have just discussed. That such a condition is necessary is shown 
by the following words: bug /b5g/, 6ee /bfy/, beetle /bfytil/, 
butterfly /b5torfl&y/. It seems ridiculous to suggest that since these 
all include /b/ and all mean some kind of insect, /b/ must be a 
morpheme. But this is only because, as native speakers, we know 
that /og/, /iy/, /iytil/, and /otorfiay/ do not exist as morphemes 
that can be associated with these words. Finally, it is necessary to 
ascertain that what we have isolated are actually single morphemes 
rather than combinations. (See 5.5.) The procedures for carrying 
out such an analysis will be sketched in the next two chapters. 

6.14 When a person is dealing with his own native language, 
much of this seems superfluous. This is simply because such com¬ 
parisons have been made repeatedly and subconsciously, if not 
consciously, in the past. We can identify English morphemes with¬ 
out detailed comparison because we have already identified most 
of them. That this is true, even of young children, can be seen from 
a common type of mistake. The child hears and learns to associate 
show /§6w/ with showed /S6wd/, tow /t6w/ with towed /t6wd/, 
etc. Then he assumes go /g6w/ must be associated in the same way 
with /g6wd/. He is, of course, wrong in detail, but right in prin- 

The Morpheme 57 

ciple, and has obviously made a morphemic analysis. He must 
merely learn the limits within which the pattern he has discovered 
is valid. 

5.15 Certain constructions composed of morphemes have a 
rigidly fixed order. For example re-con-vene (the hyphens merely 
separate the morphemes) is a familiar English word. But *con- 
re-vcne or *re-venc-con are not. They are not only unfamiliar in 
sound and appearance, but also are actually meaningless to a 
native speaker. The meaning of a word depends not only upon the 
morphemes that are present but also on the order of their oc¬ 

Other constructions allow some, but only partial, freedom of 
order. Then I went, and I went then, are both possible and have at 
most only slight difference of meaning. But *1 Vent then /. is un¬ 
intelligible because it departs from established English structure. 
In general, the more intimate constructions, like words, have the 
most rigidly fixed order, and the less closely knit constructions, 
like sentences, allow more freedom. But even longer sequences have 
some definite restrictions on order, sometimes of a subtle sort. 
For example, John came. He went away, might imply that John 
did both. But He came. John went away, certainly could not have 
that meaning. A specific reference to a person must precede a 
pronoun reference to the same person, unless some special device 
is used. This is a peculiarity of English structure, not of logic, nor 
of the general nature of speech, since some other languages have 
quite different rules. 

6.16 The fixed order of morphemes in certain constructions, 
and the definable degree of freedom, are basic to language. They 
are expressions of the systematic structure which is the real 
essence of speech. It is the business of linguistic science to describe 
these principles of arrangement in the most comprehensive and 
concise way possible. Such a description is the grammar of the 
language. The term is in poor repute with some, largely because 
of lack of precision in its use, and because it has frequently served 
as a label for legislation as to how a language should be used, rather 
than as a description of how it actually is used. These implications 
are not, of course, inherent in the term, but it is necessary to take 
care to avoid them. As used in this book, grammar will compre¬ 
hend two convenient, but not precisely delimitable, subdivisions: 

58 The Morpheme 

morphology, the description of the more intimate combinations of 
morphemes, roughly what are familiarly called "words”; and 
syntax, the description of larger combinations involving as basic 
units the combinations described under the morphology of the 
language. Some linguists use the term morphology to cover both 
subdivisions, in which case it is equivalent to grammar as used 

5.17 The grammar of a given language cannot conveniently be 
stated in terms of the arrangement of specific morphemes, because 
the total number of morphemes in any language is far too large 
to permit this. However, it is always found that the morphemes 
can be grouped into certain classes, each with a characteristic 
distribution. The structure of utterances in the language can 
then be stated in terms of these classes of morphemes. In this way 
the material which must be described is reduced to manageable 

For example, walk, talk, follow, call, etc., form an extensive class 
of morphemes. So likewise s (marking the third person singular), 
ed and ing form a smaller class. The latter can occur only imme¬ 
diately following one of the former (or some equivalent con¬ 
struction). The members of the first group can be found imme¬ 
diately preceding one of the second group, or they may be found 
alone. That is, walks, walked, walking, and walk all occur. But in 
*swalk or *ingwalk the order is wrong and the forms are accordingly 
impossible. “Walkeding is unintelligible because ing cannot follow 
ed. *Shelfed is not found because shelf belongs to another class 
which never precedes ed. All such facts, and many more like them, 
can be comprehended in a relatively few simple statements about 
the classes of morphemes. The complete listing of all possible and 
impossible sequences, on the other hand, even within a closely 
restricted sample of English, would be cumbersome and rapidly 
becomes utterly impossible as the number of morphemes treated 

5.18 The broadest and most comprehensive classes of mor¬ 
phemes in English, and the most nearly universal in the languages 
of the world, are roots and affixes. Walk, talk, follow, etc., is one 
class of roots. Shelf, rug, road, etc., is another. The vast majority of 
English morphemes are roots, and the number runs into many 
thousands. Such morphemes as -s, -ed, -Ing, etc. are affixes. Here- 

The Morpheme 59 

after affixes will ordinarily be cited with hyphens to indicate the 
manner in which they are affixed. 

A definition of these two classes which would be universally 
applicable would be immensely complex and is probably un¬ 
necessary here. A definition which will fit the needs of one specific 
language is commonly feasible. In general, affixes are subsidiary 
to roots, while roots are the centers of such constructions as 
words. Roots are frequently longer than affixes, and generally 
much more numerous in the vocabulary. 

6.19 Two different types of affixes can be defined here. Both 
are found in English and in many other languages. Prefixes are 
affixes which precede the root with which they are most closely 
associated. Examples are: /priy-/ in prefix, /riy-/ in refill, and 
/iij-/ in incomplete. Prefixes are also common in many other lan¬ 
guages. Hebrew examples are /bo-/ ‘in’ in /bobayit/ ‘in a house’ 
and /hab-/ ‘the’ in /habbayit/ ‘the house’; compare /bayit/ 
‘house.’ Suffixes are affixes which follow the root with which they 
are most closely associated. Some English examples are: /—*z/ in 
suffixes, /-iq/ in going, and /-is/ in boyish. Suffixes are also com¬ 
mon in many other languages. Some Swedish examples are ~en 
‘the’ in dagen 'the day’ and -ar ‘plural’ in dagar ‘days’; compare 
dag ‘day.’ 

Note that in English many speakers have both a prefix /iq-/ 
and a suffix /-iij/. Both may occur with the same morphemes. 
Incomplete /iqkompllyt/ and completing /kompllytiij/ arc how¬ 
ever clearly different. The position of these affixes in the word dis¬ 
tinguishes between them unquestionably. 

6.20 Affixes may be added directly to roots, or to constructions 
consisting of a root plus one or more other morphemes. All these 
may be called stems. A stem is any morpheme or combination of 
morphemes to which an affix can be added. The English word 
friends /fr6ndz/ contains a stem /frend/ which is also a root, and 
an affix /-z/. Friendships /fr£nd$ips/ contains an affix /-s/ and a 
stem /frdndsip/, which, however, is not a root since it consists of 
two morphemes. Some stems or words contain two or more roots, 
and are said to be compound. Blackbird /blsfekbSrd/ is a compound 
word, containing two roots, /blfek/ and /bord/. Blackbirds con¬ 
tains a compound stem and an affix. 

6.21 In some languages, certain affixes function primarily to 

60 The Morpheme 

form stems, and as such have little meaning other than this 
linguistic function. Such morphemes may be called stem-forma- 
tives. The Greek word /thermos/ ‘warm’ consists of a root 
/therm-/, a stem-formative /-o-/, and a final affix /-s/. The 
latter, which indicates among other things that the word can be 
the subject of a sentence, cannot be attached directly to the root. 
That is, /'therms/ is impossible. Stem-formatives of this kind are 
very common in Greek. 

Greek compound words are usually formed by compounding 
stems rather than roots. English words derived from Greek, or 
formed on the Greek pattern, commonly lose or distort suffixes at 
the end of the word, but the stem-formative of the first stem is 
usually quite evident. Thermometer is composed of the stems 
thermo- and meter. The first of these is formed from the root therm- 
bv adding the stem-formative -o-. This accounts for the very 
common occurrence of -o- in words of this type. Compare 
morph-o-logy, ge-o-graphy, philosophy , etc. 

At this point a word of warning is in order: -o- is a morpheme 
in English, not because it is one in Greek, but because certain facts 
of English structure require it to be so interpreted. We cannot be 
satisfied with dividing thermometer as either thermo-meter or as 
therm-ometer. Comparison of isotherm indicates that tlusrm- is a 
morpheme. Meter can stand alone as a word. Therefore neither 
thermo- nor -ometer is a single morpheme. The Greek origin of the 
morpheme -o- may contribute to an understanding of the history 
of this English feature, but is not otherwise relevant to English 

structure. ... 

6 22 Some morphemes have a single form m all contexts. 
English /->>)/ in coming, walking, etc,, is an example. (The fact 
that some speakers pronounce this /-in/ or /-iyn/ does not alter 
the matter. Typically, a person pronounces -ing the same in all 
linguistic contexts regardless of which pronunciation he uses.) In 
other instances there may be considerable variation. The plural 
-s is pronounced in three different common ways: in boys /boyz/ 
it is /-z/; in cals /k‘Ms/ it is /-s/; and in roses /rdwziz/ it is /-**/. 
In spite of this difference in form, every native speaker would be 
quite certain that these arc in some sense the same thing. The 
impression of the native speaker is corroborated by an examination 
of the way in which these three arc used. Inspection of large vol- 

The Morpheme 61 

umes of material will reveal that /—iz/ occurs only after /s z § t 6 ]/ 
and that neither of the others ever occurs in this place, /-s/ occurs 
only after /p t k f 0 / and neither of the others ever occurs here, 
/-z/ occurs after all other consonants and all vowels. The selection 
of the correct one of these is purely automatic with the native 
speaker, and he very rarely makes any mistake; indeed, it requires 
conscious effort to contravene these patterns. It does not matter 
whether the native speaker has ever heard the word before. Given 
a phrase like two taxemes, most people will read it/tffw+tifeksiymz/. 
But though not all agree in their guess as to how to pronounce 
the stem taxeme, all will agree in pronouncing -s in this case as 

To provide for description of such cases (and they are frequent), 
linguists distinguish between allomorphs and morphemes. An 
allomorph is a variant of a morpheme which occurs in certain 
definable environments. A morpheme is a group of one or more 
allomorphs which conform to certain, usually rather clearly de¬ 
finable, criteria of distribution and meaning. Thus /-z/, /-s/, and 
/-iz/, above, are three allomorphs of a single morpheme. They are 
such because they show the definite definable distribution which 
was mentioned, and because they have the same meanings. 

5.23 The concept of allomorphs and morphemes, and of other 
“alios” and “ernes,” is one of the most basic in descriptive lin¬ 
guistics. Its importance both as a tool and as an insight into the 
operation of language can hardly be overestimated. It stands 
behind the two basic units of linguistic description, the phoneme 
and the morpheme, as well as behind other lesser concepts such 
as the grapheme. The principle involved is largely responsible for 
the high development of linguistic theory and techniques. The 
inapplicability (so far as we now know) of the concept in certain 
related disciplines is the chief differentiating factor between the 
science of linguistics and other treatments of human behavior. 

6.24 Any phenomenon is said to be conditioned if it occurs 
whenever certain definable conditions occur. This is not identical 
with saying that it is caused by these conditions. All that is implied 
is that they occur together in some way, so that one can be pre¬ 
dicted from the other. Where there's S7twke there's fire, and Where 
there's fire there’s smoke, are both statements of conditioning. Only 
one of them can possibly be a statement of cause, and there is no 

62 The Morpheme 

need to assume that either necessarily is. The three allomorphs of 
the plural morpheme, /-z/, /-s/, and /—iz/, are conditioned, since 
each occurs when certain clearlj r defined conditions occur. In this 
case the conditioning factor is the phonetic nature of the preceding 
phonemes, /-z/ occurs only after voiced sounds; /-s/ only after 
voiceless sounds; and /-iz/ only after groove fricatives and affri¬ 
cates. We may therefore say that they are phonologically condi¬ 
tioned. This means that, if we understand the facts of distribution, 
we can accurately predict which of the three will occur in any place 
where any one of them could occur. As native speakers of English 
we make this selection automatically and subconsciously. As lin¬ 
guists we may formulate a descriptive statement of our own habits 
and on the basis of this statement make the proper selection 
deliberately. The formal statement is valid only insofar as it pro¬ 
duces the same result as the subconscious habit of the native 

This automatic selection is part of the structure of English and 
has to be learned. It is not “just natural,” even though it may 
seem so to us. To a foreigner it may seem very unnatural. Indeed, 
it is not a universal feature of English. In the Blue Ridge Moun¬ 
tains of Virginia /-iz/ is used not only after /s z 5 l fi }/ but also 
after /spstsk/. Thus wasps, posts, and tasks are pronounced 
/waspiz/, /pdwstiz/, and /txfcskiz/, not /wasps/, /powsts/, and 
/tefcsks/ as in most dialects. In both dialects the form is phono¬ 
logically conditioned; in both the selection is completely auto¬ 
matic and quite regular. They are merely different, and each seems 
entirely natural to the speakers. 

6.26 The selection of allomorphs may also be morphologically 
conditioned. In this case the selection is determined by the specific 
morpheme or morphemes forming the contexts, rather than by any 
phonologic feature. The plural of ox is oxen /&ksin/. /-in/ is an 
allomorph of the plural morpheme which is used only with this 
one root /aks/. For the native speaker familiar with the word (a 
dwindling minority but for its use as a grammatical example!), 
/-in/ is automatically selected after /aks/ and /*&ksiz/ is rejected 
as incorrect. There is nothing phonologic about this selection. 
Boxes, foxes, axes are phonologically similar, but use /-iz/. The 
peculiarity rests in the morpheme /aks/ as a morpheme; the selec¬ 
tion is morphologically conditioned. 

The Morpheme 63 

5.26 The concept of allomorph and morpheme necessitates 
some additional notation to avoid long circumlocutions. Variation, 
as that between allomorphs, will be indicated by the sign to be 
read as “varies with” or “alternates with” or simply “or.” Wo 
may thus write /-z ~ -s ~ —iz/ to indicate that these three are 
to be taken as allomorphs of one morpheme. The same symboliza¬ 
tion can be used to identify the morpheme. 

If the morpheme has numerous allomorphs, as many do, it is 
awkward to have to list all of them every time the morpheme is 
mentioned. Instead, it is desirable to have a single symbol to 
indicate a morpheme as such, comprehending all the variant forms 
in which it can appear. For this purpose we use braces { }. Within 
the braces we may place any convenient designation of the mor¬ 
pheme at hand. Of course, once such a symbol has been defined, 
we arc no longer at liberty to choose arbitrarily. For example, wo 
could choose J-s) or |-z| or various other designations for the 
English plural morpheme. However, to conform to one established 
notation we will select the symbol |—Zi}. Having defined this 
symbol as equivalent to /-z ~ -s ~ -iz ~ -in we will not 

again need to specify the allomorphs included. {—Z»} will be read 
as “the morpheme Z one.” 

5.27 It will be convenient at this place to recapitulate the 
various types of notation which we will use: 

[ ] indicate a phonetic transcription, in which the pronuncia¬ 
tion is transcribed as heard, not necessarily representing the 
significant features. 

/ / indicate a phonemic transcription in which the pronuncia¬ 
tion is transcribed so as to represent all significant features 
and nothing else. 

{ } indicate a morphemic representation in which one arbi¬ 
trarily selected symbol is used to represent each morpheme 
and comprehend all its allomorphs. It does not directly give 
any information about pronunciation. 

Italics indicate orthography, the traditional spelling which may 
approach being phonemic or may give almost no direct infor¬ 
mation about pronunciation. The relationship between 
spelling and pronunciation varies from language to lan¬ 

64 The Morpheme 

* * indicate glosses, translations, or other indications of the 
meaning of items. 

* indicates that a form is impossible or unknown. 

The uses of [ ] and / / are well established and used by the vast 
majority of American linguists. { ) is used by a considerable 
group, but other notations are also in vogue. There are no uni¬ 
versally established conventions for citing orthography or glosses, 
but those used here are not uncommon. 



The Identification of 

6.1 The investigation of the grammar of any language proceeds 
from the examination of a corpus, a sample of utterances which 
have been gathered for the purpose of analysis. It will seldom be 
possible to recognize the morphemes within this corpus by any 
simple inspection. The task of the investigator is, therefore, to 
select out of the corpus those pairs or sets of utterances which can 
most profitably be compared, to draw the most reasonable deduc¬ 
tions from them, and to build these together into an integrated 
and consistent system that will account for the data in the corpus. 
This analysis is then cheeked against additional utterances. On the 
basis of this increased corpus it may be corrected and extended 
until it seems probable that the results will apply usefully to any 
sample of the language. 

The first objectives are: to segment the corpus — that is, to 
divide it into portions, each of which represents a single morpheme; 
and to class these segments together into morphemes. Unfor¬ 
tunately, there is no procedure which will lead automatically to a 
correct segmentation or classing unless, like some of the elementary 
problems in the workbook accompanying this text, the corpus is 
highly artificial in selection or arrangement. False starts are almost 


66 The Identification of Morphemes 

inevitable. Some of the most promising comparisons may lead to 
nothing useful. The most significant contrasts may be difficult to 
see at first. Very frequently the several aspects of the problem are 
so interrelated that many tentative decisions must be made before 
any of them can be subjected to a real test. It is, therefore, essential 
that the analyst should bear in mind the tentative nature of all 
his results until many lines of evidence can be made to converge 
in support of each, and until they can be fitted together into a 

6.2 There are several levels of structure in a language. For the 
most useful description, each must be clearly distinguished from 
all the others. They are not, however, wholly independent. Each 
higher level of structure is best stated in terms of the units of the 
preceding level. Thus, the statement of the morphemes is best 
made in terms of the phonemes. But, on the other hand, the 
phonology should be stated without any reference to the grammar. 
So likewise, the syntax must be stated in terms of the morpheme 
sequences described in the morphology, but the description of the 
morphology must not be dependent upon that of the syntax. 

It is not ordinarily feasible to analyze each level separately. 
Instead, the work must be carried on more or less simultaneously 
on all levels. All that is required is that it be done in such a way 
that the results can be stated in terms of an orderly hierarchy of 
levels, each dependent on those below and describable without 
dependence on any above. 

The processes of linguistic analysis, though carried on more or 
less together, must be described separately. In this textbook it has 
seemed best to present grammar before phonology. We must, 
therefore, assume that at least a preliminary analysis of the 
phonology has been made. The examples in the text and the prob¬ 
lems in the workbook are all either in a phonemic transcription, or 
in some notation which is nearly phonemic. 

6.3 The identification of morphemes is done almost wholly by 
variations and refinements of one basic technique. This is the com¬ 
parison of pairs or sets of utterances which show partial contrast 
in both expression and content. Unless the contrast is partial (that 
is, unless there is some apparent identity somewhere in the utter¬ 
ances), and unless it exists in both expression and content, the 
comparison is fruitless. In many important respects this is the 

The Identification of Morphemes 67 

same general procedure as was used in Chapters 2, 3, and 4 to 
identify English phonemes. For the identification of phonemes we 
wanted the smallest possible difference in expression with any 
difference in content whatever. For the identification of mor¬ 
phemes wc seek the smallest differences of expression which exist 
with a partial difference of content. This difference of pro¬ 
cedure rests in the fundamental difference between phoneme and 
morpheme. The phoneme is the smallest significant unit in the 
expression which can be correlated with any difference in the 
content structure. The morpheme is the smallest significant unit 
in the expression which can be correlated with any one par¬ 
ticular difference in the content structure. 

Because we cannot tell what are, properly speaking, morphemes 
until wc have nearly completed the task of analysis, it will be 
permissible to use the term morpheme somewhat loosely. We must 
gradually become more precise in our use of it, however, as we 
attain to a more precise knowledge of the structure. Above all, we 
must be aware of the tentative nature of all our preliminary 
findings and never allow ourselves to be misled by appearances of 

6.4 The process of analysis is best shown by detailed discussion 
of an actual example. For this purpose we will use a scries of 
Hebrew verb forms. The data will be introduced a few words at a 
time. This is an artificial feature of the presentation. The pre¬ 
ceding step is merely implied: namely that wc have selected from 
the corpus those pairs or sets of items that can bo profitably com¬ 
pared. The order of presentation is not necessarily that which is 
most efficient for the analysis of the data, but that which most 
effectively illustrates the methods used. 

1. /zokartfihuu/ ‘I remembered him’ 

2. /zokartlihna/ ‘I remembered her’ 

3. /zokartlikaa/ ‘I remembered thee* 

Comparison of items 1 and 2 reveals one contrast in expression, 
/-uu/ : /-aa/, and one in meaning, as shown by translation, and 
hence presumably in content ‘him’ : ‘her.’ This may (tentatively!) 
be considered as a pair of morphemes. However, comparison of 
1, 2, and 3 suggests that the first identification was wrong. The 
contrast now seems to be /-huu/ ‘him’ : /-haa/ ‘her’ : /-kaa/ 

68 The Identification of Morphemes 

‘thee.’ We can be reasonably sure that the morpheme meaning 
‘him’ includes the sounds /-uu/ or /-huu/, but until we can 
identify the remaining parts of the word we cannot be sure how 
much else is included. 

6.5 4. /zokarnuuhuu/ ‘we remembered him' 

5. /zokarnduhaa/ ‘we remembered her’ 

6. /zakamuukaa/ ‘we remembered thee’ 

Comparison of 4, 5, and 6 with 1, 2, and 3 reveals a contrast in 

expression and meaning between /—tii—/ ‘I’ and /-ntiu-/ ‘we.’ 
However, as before, we cannot be sure how much is to be included 
until the remainders of the words arc identified. It is conceivable 
that the morphemes might be /—rtfi—/ ‘I’ and /-rniiu-/ ‘we.’ In¬ 
cidentally, comparison of 4, 5, and 6 with one another confirms 
our conclusion in paragraph 6.4. 

6.6 7. /qotaltfihuu/ ‘I killed him’ 

8. /qotalmiuhuu/ ‘ we killed him ’ 

Comparison of 7 and 8 with the foregoing gives us a basis for 
identifying /zakar-/ ‘remembered’ and /qotal-/ ‘killed.’ By so 
doing we have tentatively assigned every portion of each word to 
a tentative morpheme. We have, however, no reason to be certain 
that each portion so isolated is only a single morpheme. We have 
only reasonable assurance that by dividing any of these words in a 
manner similar to /zokar-tli-huu/ we have divided between 
morphemes, so that each piece consists of one or more essentially 
complete morphemes; that is, each piece is probably either a mor¬ 
pheme or a morpheme sequence. 

6.7 The problem is somewhat simpler if one sample is identical 
with another except for an additional item of meaning and of 

/koohgen/ ‘a priest’ 

/lokoohgen/ ‘to a priest’ 

There can be little doubt as to the most likely place to divide, and 
we can be rather confident in identifying two tentative mor¬ 
phemes /la-/ ‘to’ and /kooh^en/ ‘priest.’ Nevertheless, there are 
significant possibilities of error, so that this sort of division must 

The Identification of Morphemes 69 

also be considered tentative. Consider the following English 

/him/ ‘a song used in church’ 

/hfmnol/ ‘a book containing /hlmz/’ 

The obvious division is into two morphemes /him/ and /-nol/. 
Reference to the spelling (which is, of course, never conclusive 
evidence for any thing in spoken language!), hymn : hymnal sug¬ 
gests that this is not very certain. Actually the two morphemes 
arc /him ~ himn-/ and /—ol/, as may be shown by comparing 
additional data: confession : confessional, hymnology : geology, 
hymnody : psalmody. 

6.8 9. /zokaaruuhuu/ ‘they remembered him’ 

10 . /zokaarAthuu/ ‘she remembered him’ 

If we compare 9 and 10 with the foregoing we find /-huu/ 'him/ 
/-du-/ 'they/ and /-fit-/ ‘she/ But where 1-6 have /zokar-/, 9 
and 10 have /zokaar-/. There is an obvious similarity of form 
between /zokar-/ and /zokaar-/ and the meaning seems to be 
identical. We may guess that they are two different allomorphs of 
one morpheme, and proceed to check whether this hypothesis is 
adequate. The method of doing so will be discussed in Chapter 7. 
Wo must leave the question until that time, hut must anticipate 
the result, /zokar-/ and /zokaar-/ will be shown to be variants of 
one tentative morpheme. 

But though we will proceed on the basis that the hypothesis can 
be sustained, we must recognize that there are certain other 
possibilities. 1) /zokar-/ and /zokaar-/ may bo different mor¬ 
phemes. This seems unlikely because of the similarity of meaning, 
but we must always remember that English translation may be 
misleading. 2) A somewhat less remote possibility is that /zokar-/ 
and /zokaar-/ are each sequences of morphemes and contain two 
contrasting morphemes. We can do nothing with this possibility 
from the data at hand, because there is no evidence of a contrast 
in meaning, but this may well be the kind of difference that does 
not show up clearly in translation. 3) We may have divided 
wrongly. Perhaps 'I' is not /-tfi—/ but /—a—tfi—/ and ‘they’ is 
similarly /-aa-du-/. This would mean that the morpheme for 
‘remembered’ would have to be /zok-r-/. Our only present reason 

70 The Identification of Morphemes 

for rejecting this possibility is the comparative rarity of discon¬ 
tinuous morphemes. We would ordinarily assume that morphemes 
are continuous sequences of phonemes unless there is cogent reason 
to believe the contrary. 

6.9 11. /zokartuunii/ ‘you remembered me’ 

We have as yet no item which forms a wholly satisfactory com¬ 
parison with 11. We may, however, tentatively divide it into 
/zokar-/ + /-ttiu-/ ‘you’ 4- /—nii/ ‘me.’ We do this because we 
have come to expect words similar to this to be divisible into three 
pieces, stem + actor + person acted upon, in that order. A division 
on such a basis is legitimate if done with caution, though obviously 
such an identification is not as certain as it would be if based on 
contrasts for each morpheme separately. 

6.10 12 . /fsomarttiuhaa/ ‘you guarded him’ 

13. /loqaaxuunii/ ‘they took me’ 

Even without providing minima! pairs, 12 and 13 pretty well 
corroborate the conclusion which was drawn from 11 in para¬ 
graph 6.9. They thus confirm the two morphemes /-tuu-/ ‘you’ 
and /-nii/ ‘me.’ Words 11, 12, and 13 would be rather unsatis¬ 
factory words from which to start an analysis. However, as the 
analysis proceeds, the requirements for satisfactory samples relax 
in some respects. This is because we are now able to make our 
comparisons within the framework of an emerging pattern. This 
pattern involves certain classes of elements, stems, actor affixes, 
and affixes stating the person acted upon. It involves certain 
regular types of arrangement of these elements. In short, the 
pattern we are uncovering is a portion of the structure of the 
language at a level a bit deeper than mere details of individual 

6.11 14. /zokaar6o/ ‘he remembered him’ 

This word cannot be analyzed by comparison with the foregoing 
only. We can easily identify the stem as /zokaar-/, identical in 
form with that of 9 and 10. But the remainder /-6o/ neither seems 
to consist of the expected two parts (actor and person acted upon), 
nor to contain the morpheme /-huu/ ‘him’ which meaning would 


The Identification of Morphemes 

lead us to expect. Since the pattern does not assist us here in the 
way it did with 11, we must seek some more direct type of evidence. 

6.12 15. /zaak&rtii/ ‘I remembered' 

16 . /zaak&rnuu/ ‘ we remembered ’ 

17. /zaak&r/ ‘he remembered’ 

These three forms differ from all those examined before in that 
they do not express a person acted upon. If we compare these words 
with each other, and if we compare 15 and 16 with 1 and 4, we 
can easily identify the affixes expressing the actor. These are 
/—tii/ ‘I’ and /-nuu/ ‘we,’ identical with those we found before 
except for a difference in the stress. In 17, however, there is no 
affix expressing actor. We will tentatively list 0 (zero) ‘he’ with the 
other actor affixes. This is intended merely as a convenient nota¬ 
tion for our conclusion that the uctor ‘he’ is expressed by the 
absence of any affix indicating some other uctor. These three forms 
also show another variant of the stem: /zaakar/; we shall proceed 
on the hypothesis that like /zokar-/ and /zokaar-/, it is merely 
another conditioned variant. This proposal should be carefully 
checked by methods to be discussed later. 

6.13 The analysis attained in the last paragraph suggests that 
item 14 can be considered as divisible as follows: /zokaar-0-5o/. 
The zero is, of course, a fiction, but it does serve to indicate that the 
form does show a rather closer parallelism with the others than wo 
could see at. first. That is, it contains a stem and a suffix expressing 
the person acted upon, and these are in the same order that we 
have found before. Whereas the pattern we had found did not seem 
to fit this word, closer examination shows that it does fit in with 
only slight modification. The pattern is therefore valid. 

One problem posed by item 14 is taken care of in this way, but 
the other remains. We have identified two forms meaning ‘him,’ 
/-huu/ and /- 6 o/. These are not so obviously similar in form as 
/zokar-/ and /zokaar-/, so the hypothesis that they arc allomorphs 
of one morpheme is not so attractive. Nevertheless, the similarity 
in meaning, and certain peculiarities in distribution which would be 
evident in a larger body of data, should induce us to check such a 
hypothesis. It will be sustained; /-huu ~ - 60 / is one morpheme. 

6.14 In the course of the discussion we have found four stems: 
/zokar-/ ‘remembered,’ /qotal-/ ‘killed,’ /somar-/ ‘guarded/ 

72 The Identification of Morphemes 

and /loqaax-/ ‘took.’ Comparison of these forms reveals that they 
all have the same vowels and differ only in consonants, /loqaax-/ 
is not an exception, since it compares directly with /zokaar-/. 
More data would yield a much longer list of such forms. This 
similarity in vowels could be a coincidence, but that possibility is 
slight. Another hypothesis is that these forms consist of two mor¬ 
phemes each. This is very attractive, but there is no means of 
checking it without a contrast. The following will provide such: 

18 . /Soom^er/ ‘watchman’ 

19. /zookger/ ‘one who remembers’ 

20 . /qoot^el/ ‘killer’ 

By comparing these with some of the earlier samples we may 
identify the following morphemes: /z-k-r/ ‘remember,’ /q—t—1/ 
‘kill,’ /§-m-r/ ‘guard,’ /1-q-x/ ‘take,’ /-oo-4e-/ ‘one who,’ and 
/-o-a- ~ -o-aa- ~ -aa-a-/ 1 -ed.’ The first four of these are 
roots; the last two are some sort of affixes. 

6.15 Note that we were wrong in considering /zokar-/, 
/zakaar-/, and /zaakar/ as allomorphs of a single morpheme. No 
damage was done, however, since these three forms, each composed 
of two morphemes, are distributed in exactly the same way as are 
allomorphs. What we assumed to condition the selection of one of 
these three (/zakar-/ etc.) can just as well be considered as con¬ 
ditioning the selection of one of the allomorphs of the affix con¬ 
tained in these stems. Treating larger items as morphemes is, of 
course, wrong, but not seriously so at preliminary stages, provided 
the larger units consist of associated morphemes. Ultimate sim¬ 
plification is, however, attained by full analysis in any case like 
that just discussed. 

6.16 That the analysis in paragraph 6.14 should yield mor¬ 
phemes such as /z-k-r/ and /-oo-6e-/ seems at first sight some¬ 
what disconcerting. We expect morphemes to be sequences of 
phonemes. These, however, are discontinuous and interdigitated. 
Of course there is no reason why such morphemes cannot occur, 
as in fact our sample has indicated they do. They are much less 
common than compact sequences of phonemes, but they occur in 
a wide variety of languages and are quite common in some. Any 
combination of phonemes which regularly occur together and 
which as a group are associated with some point in the content 

The Identification of Morphemes 73 

structure is a morpheme. We need give no regard to any peculiarity 
of their arrangement relative to each other and to other phonemes. 
Rarely do morphemes consist of separate portions widely separated 
by intervening material. A linguist must always be prepared for 
such a phenomenon, however, rare as it may be. 

Hebrew and related languages are unusual in the large number 
of discontinuous morphemes they contain. In fact the majority of 
the roots are similar to {zkr}, consisting of three consonants. 
Various allomorphs occur: /z-k-r/ in /zaakdr/ ‘he remembered,’ 
/-zk-r/ in /yizk6or/ ‘he was remembering,’ and /z-kr-/ in 
/zikrli/ ‘my remembrance.’ The three consonants never occur con¬ 
tiguously in any utterance; such roots are discontinuous in all their 

6.17 In other languages, discontinuous allomorphs of otherwise 
quite usual morphemes occur. These commonly arise as a by¬ 
product of a special type of affix not mentioned before, an infix. 
An infix is a morpheme which is inserted into the stem with which 
it is associated. In comparison with suffixes and prefixes, infixes 
are comparatively rare but of sufficiently frequent occurrence to 
warrant notice. An example is the common Greek stem formative 
/-m-/ in /lambano-/ ‘I take’ from the root /lab-/- Another is 
Quileute (Oregon) /-£-/ ‘plural’ in /ho^k w at’/ ‘white men’ from 
/hok' v at’/ ‘white person.’ Such infixes produce discontinuous 
allomorphs /la-b-/ and /ho-k w at’/ of the root morphemes with 
which they occur. 

6.18 An affix should not be considered as an infix unless there 
is cogent reason to do so. Of course, any affix which actual I}' 
interrupts another morpheme is an infix. In Tagalog ginulag 
‘greenish blue’ is formed from the root gulag ‘green vegetables.’ 
The -in- is clearly an infix. But it is not justifiable to consider 
English -as- in reassign as in infix. This word is made by two pre¬ 
fixes. First as- and sign form the stem assign. Then re- is added. 
The alternative would be to consider rc- and sign as forming a 
stem resign to which an infix -as- is added. The latter would be 
immediately rejected by any native speaker of English, since he 
would sense that reassign has a much closer connection with assign 
than with resign. It is always better, unless there is good reason 
to the contrary, to consider words as being constructed of suc¬ 
cessive layers of affixes outward from the root. 

74 The Identification of Morphemes 

6.19 Most English verbs have a form that is made by the 
addition of the suffix -ed /-d ~ -t ~ -id/. This is usually known as 
the past. The verbs which lack this formation do, however, have 
some form which is used in all the same syntactic environments 
where we might expect such a form, and in comparable social and 
linguistic contexts. For example, in most of the places where dis¬ 
cover /disk5vor/ can be used, find /faynd/ can also. Similarly, 
where discovered /diskSvord/ can be used, found /fawnd/ gen¬ 
erally can also. Found must therefore be considered as the past of 
find in the same sense that discovered is the past of discover. 

Most of the past tenses which lack the -ed suffix are clearly 
differentiated from the base form by a difference of syllable nucleus. 
We may express the facts by the following equations: 

discovered = discover + suffix -ed 

found = find + difference of syllable nucleus 

When it is so stated, it becomes evident that the difference of syl¬ 
lable nucleus functions in some ways like the suffix. We may con¬ 
sider such a difference in phonemes (they are not restricted to 
nuclei; consider send : sent) as a special type of morphemic element 
called a replacive. 

We will use the following notation for a replacive: /aw <— (ay)/. 
This should be read as “/aw/ replaces /ay/.” The equation above 
can be stated in the following form: 

found = find + <w«— (t) 

/faund/ = /fHynd/ + /aw«— (ay)/ 

If this is done, then we must consider /aw ♦— (ay)/ as another 
allomorph of the morpheme whose most familiar form is -ed and 
which we can conveniently symbolize {-Dj}. This morpheme has a 
number of replacive allomorphs, some of which will be listed in 
8.10. All of them are morphologically conditioned. the 

English noun plural affix, also has replacives among its allomorphs. 

6.20 It is, of course, possible to describe a language like English 
without recourse to replacives. Thus, geese /giys/ can be described 
as containing a root /g-s/ and an infix allomorph of the plural 
morpheme [-Zi } of the form /—iy—/. Then the singular would have 
to be described as containing an infix /-uw-/, an allomorph of a 
singular morpheme *{X}. Except for the cases under consideration, 

The Identification of Morphemes 75 

there are no infixes, nor discontinuous morphemes in the language. 
To consider plurals like geese as formed by an infix turns out to 
involve many more complications than the alternative of de¬ 
scribing replacives. As is often the case, the simpler explanation 
accords more closely with the native speaker’s feeling about his 

6.21 With replacives it is not easy to divide a word into its 
constituent morphemes. Obviously /giys/ is two morphemes, but 
the four phonemes cannot be neatly apportioned between them. A 
morpheme does not necessarily consist of phonemes, but all mor¬ 
phemes are stateable in terms of phonemes. A replacive must be 
described in terms of two sets of phonemes: those that appear 
when it is present (/iy/ in geese) and those that appear when the 
replacive is absent (/uw/ in goose). A morpheme can consist of any 
recurring feature or features of the expression which can be de¬ 
scribed in terms of phonemes, without restriction of any sort. 

6.22 A further, and in some respects more extreme, type of 
morphemic element can be seen in the past of some other English 
verbs. Words like cut and hit parallel such forms as walked in 
meaning and usage. There is, however, no phoneme difference of 
any kind between the past and the non-past form. Nevertheless, 
it is in the interests of simplicity to consider all English past verb 
forms as consisting of a stem plus an affix. Moreover, the descrip¬ 
tion must in some ways note the lack of any overt marker of the 
past. An expedient by which both can be done is to consider cut 
‘past’ as containing a root /kot/ plus a zero affix. (Zero is cus¬ 
tomarily symbolized 0 to avoid confusion with the letter O.) 
0 is therefore another of the numerous allomorphs of {-Dij. 

The plural affix {-Zij also has a zero allomorph in sheep. The 
reason that it is necessary to describe these forms in this way rests 
ultimately in English content structure. Native speakers feel that 
the dichotomy between singular and plural is a basic character¬ 
istic of nouns. Every individual occurrence of any noun must be 
either singular or plural. Sheep is ambiguous, but not indifferent 
to the distinction. That is, in any given utterance the word is 
thought of by speaker and hearer as either singular or plural. 
Sometimes they may disagree, plural being intended and singular 
being perceived, or vice versa. It requires conscious effort for a 
person accustomed only to English patterns to conceive of noun 

76 The Identification of Morphemes 

referents without consideration of number. To attempt to do so 
impresses many people as being “too abstract.” Yet they feel 
under no such compulsion to distinguish the exact number if it is 
more than two. 

In other words, there is a covert difference between sheep 
‘singular’ and sheep ‘plural,’ and this is linguistically significant, 
as may be seen from the fact that it controls the forms of certain 
ot her words in This sheep is .... : These sheep are ... . The recog¬ 
nition of a 0 allomorph of {-Z,| is merely a convenient device for 
entering all this into our description. 

6.23 The other possible use of the zero concept in morphology 
would be to set up a 0 morpheme, that is, one in which there is no 
overt allomorph whatever. To do so is quite unnecessary, and will 
generally lead to increased complexity of statement, the precise 
opposite of our desired goal. Moreover, it is logically indefensible. 
If we are to make such free use of zero, there is no definable place 
to stop. We could freely add zeros of all kinds to our descriptions, 
and each would be as justifiable as the last. The situation is de¬ 
cidedly different in cases in which zero is an allomorph of some 
morpheme which more commonly has an overt form. We can never 
add zeros beyond the limits of the gaps clearly visible in the 
structure being described. 

Sometimes it is convenient to use the symbol 0 as a temporary 
expedient in analysis. That was done in (5.12. Here it merely served 
as a short notation to indicate that no morpheme for the third 
person masculine singular was found. This fact demanded record 
during the course of the investigation. At a later stage some other 
less artificial method can be found to state the facts. If zero is used 
in this way as a temporary device, no harm will come; but the 
temptation to take such a zero too seriously must always be 
resisted. Unless it later proves to be an allomorph of some mor¬ 
pheme having some observable forms, then it will have to be 
eliminated in a final statement. 

6.24 In 6.5 the principle was laid down that all parts of a 
sample must be identified before it is passible to consider a division 
into morphemes as established. However, some morphemes occur 
only in a very limited distribution, so that this is not always pos¬ 
sible. In cranberry /krsenb&riy/ the division would seem to be be¬ 
tween /knen/ and /beriy/. The latter can be confirmed by numcr- 

The Identification of Morphemes 77 

ous words including berry, blueberry, blackberry, but the first 
element does not seem to occur in any other context. We arc 
justified in considering this as a morpheme only after a thorough 
analysis of a large segment of English morphology. This will reveal 
certain widespread patterns of word formation which will confirm 
a division between /knen/ and /beriy/ in the lack of any more 
direct evidence. Nevertheless, /kram/ cannot be considered as 
securely established as a morpheme as, say /beriy/, where both 
morphologic patterns and direct contrasts converge. 

6.26 Morphemic analysis is hardly practical without close 
attention to the meanings of forms. This must ordinarily be 
manipulated in the form of translations. Translation can obscure 
some features of meaning and falsify others. A contrast in meaning 
is not relevant unless there is also a contrast in form. Consider 
the following additional data for the problem which was discussed 

21 . /zokartiihuu/ ‘I remember him’ 

22 . /zokartiihuu/ ‘I will remember him’ 

The contrast in meaning between these two items and item 1 must 
be disregarded because there is no contrast in form. Actually, this 
contrast is introduced in translation. It involves a category (time 
of action) which is not expressed in the inflection of the Hebrew 
verb. It is, however, impossible to cite an English verbal phrase 
without specifying time. The category must be added in the gloss, 
even though there is no justification in the Hebrew. 

6.26 The following word may also be added to the data of the 
same problem: 

23. / 9 ezkor6chuu/ ‘I remember him’ 

There is a profound contrast in expression between 21 and 23, but 
none in meaning as shown by the translations. This form cannot 
be fully analyzed until the difference in meaning is established. The 
contrast in form is, however, much more significant than the 
apparent identity in meaning, since the latter may be purely a 
matter of translation. Finding these or similar words in actual 
contexts may supply the contrast in meaning which is lacking in 
the translation. Translation is a very inadequate means of ex¬ 
pressing meanings and must always be used with great caution. 




Classing Allomorphs into 

7.1 The analytic procedures described in Chapter 6 yield a list of 
tentative morphemic elements. Even before the procedures are com¬ 
pleted, it is necessary to begin organizing the resulting elements in 
such a way as to make clear the existing structural relationships. 
One step in organization is to determine which of the elements 
are to be classed together as allomorphs of a single morpheme. 

In the example discussed in Chapter 6, two very similar ele¬ 
ments, /zakar-/ and /zokaar-/, were isolated. These were ob¬ 
served to have similar, but not identical, phonemic form, and to 
be apparently the same in meaning. The hypothesis was suggested 
that these two are allomorphs of one morpheme, and the work was 
carried forward on this assumption. Of course, such a hypothesis 
should have received at least preliminary checking before further 
analysis was based on it. In a practical situation it would have. 
But, for clarity of presentation, the discussion of the methods used 
in such a check had to be deferred to this chapter. This is another 
instance of the fact that various stages in the analysis cannot be 
neatly separated, though logically they should follow one after 
another. To go back and forth between various steps in analytic 
procedure and to perform many of them more or less simultane- 


Classing Allomorphs into Morphemes 79 

ously is essential for efficient work. It is, however, never necessary, 
and often dangerous, to confuse them. After an analysis has been 
achieved, it should always be possible to go back to the original 
data and demonstrate all the conclusions systematically in the 
logically prescribed order. If this is not possible, that is, if the 
various levels of structure arc inextricably mixed, the whole result 
must be considered suspect. 

7.2 A morpheme is the smallest element in the expression 
which has a direct relationship with any point in the content 
system. For any two items to qualify as allomorphs of a single 
morpheme it is essential merely that both have the same relation¬ 
ships to the same structure points in the content. Unfortunately, 
this is not directly observable. The process of grouping elements 
into morphemes therefore depends on finding, through various 
subsidiary or even incidental criteria, classes of items which may 
be assumed with reasonable probability to have such relationships. 
Two observable features may be expected to show some correlation 
with the expression-content relationship which would certainly 
class elements into morphemes: these two are meaning (in the 
vague and somewhat unscientific sense) and distribution. Of these 
the most objectively observable is distribution. Nevertheless, we 
must take care to select the most pertinent features of distribution. 
When samples are small, we must be particularly cautious to dis¬ 
criminate between significant and chance occurrences. Meaning is 
subject to a different kind of difficulty. It will ordinarily have to be 
judged from translations or from the non-linguistic context. The 
latter is the continuum upon which the content structure is 
arbitrarily imposed. We have no direct way of judging the extent 
or nature of the arbitrariness, but we can be sure that it exists. 
Translation meanings, on the other hand, include such arbitrary 
structuring, but they include not only that of the language under 
consideration but that of the language of the glosses as well. 
Meaning is, therefore, a variable which is not subject to any precise 
control. It will never be safe to use it alone, but only in combina¬ 
tion with some facts of distribution. But it will not be possible to 
use distribution alone (except perhaps in special cases), since mean¬ 
ing will be needed to assess the pertinence of the distributional 
features. It follows that in each case we must use a double criterion, 
both parts of which must be satisfied. 

80 Classing Allomorphs into Morphemes 

Various types of distributional criteria must be used in different 
ways. We will therefore state and discuss separately two different 
sets of criteria for the classing of allomorphs into morphemes. 

7.3 Two elements can be considered as the same morpheme if 
( 1 ) they have some common range of meaning, and (2) they are in 
complementary distribution conditioned by some phonologic feat¬ 
ure. (Morphologic conditioning will be discussed below.) 

Two elements are said to be in complementary distribution if 
each occurs in certain environments in which the other never 
occurs — that is, if there are no environments in which both occur. 
For the present this means environments definable in terms of cer¬ 
tain phonemes, types of phonemes, or combinations of phonemes. 
Complementary distribution is one of the basic concepts of lin¬ 
guistic theory and method, and will recur repeatedly in various 
contexts in this book and all linguistic literature. It is frequently 
abbreviated CD. 

7.4 Note that the criterion of 7.3 does not include any reference 
to phonemic similarity. There is no need that allomorphs be 
similar in form. Occasionally the difference may be as great as that 
in the root of go /g6w/ and went /\v£nt/. Most often, however, 
allomorphs arc phoncmically similar. Commonly they differ by 
the smallest degree possible within the phonemic system of the 
language. For this reason, phonemic similarity is important in the 
analysis, but not as a criterion for uniting elements. Instead it 
serves primarily to indicate pairs of elements that should be tested. 
This was the case in the last chapter with /zakar-/ and /zokaar-/. 

7.6 To determine whether two elements are in complementary 
distribution, it is necessary to tabulate a considerable number of 
occurrences of each. The next step is to examine each list, looking 
for some feature in the environment in one group which is never 
present in the other. Continuing the example of 6.8 but disre¬ 
garding all later conclusions, we may tabulate the relevant forms 
as follows (the hyphens are inserted merely to separate the ele¬ 
ments under discussion from their environments): 

With short vowel With long vowel 

1 . /zokar-tlihuu/ 5. /zokar-nuuhaa/ 9. /zokaar-uuhuu/ 

2. /zokar-tfihaa/ G. /zakar-nuukaa/ 10. /zokaar-athuu/ 

3. /zakar-tiikaa 11. /zokar-tuunii/ 14. /zokaar-6o/ 

4. /zokar-nuuhuu/ 

Classing Allomorphs into Morphemes 81 

If we examine these lists, we find that /zokar-/ occurs before 
a consonant, whereas /zokaar-/ occurs before a vowel. Unfor¬ 
tunately the evidence is rather scant, so that any conclusion must 
be very tentative. Nevertheless, this will be sufficient to permit us 
to continue with the analysis as we did in 6.8. 

7.6 Cases frequently arise, such as that just discussed, in which 
the available evidence is too restricted to afford any high degree 
of certainty. In this case, even if all the occurrences of /zokar-/ 
and /zokaar-/ in a much larger corpus were to be examined, it 
would still leave much room for doubt. However, similar patterns 
in other morphemes will emerge as the work progresses. Each of 
these strengthens the case for all the others. Finally, it will be 
possible (for Hebrew) to establish a pattern of changes in vowel 
length which are repeated over and over through the language. 
This general pattern will provide the confirmation for our analysis 
which could never be obtained from the examination of one such 
case alone. The pattern includes the following generalizations 
which apply to our example: (1) Any vowel followed by two con¬ 
sonants is short unless stressed, when it may be either long or short, 
/zokartiihuu/. (2) Ordinarily any vowel followed by one consonant 
is long if the next vowel is stressed, /zokaaruuhuu/. 

Conclusions concerning individual morphemes can often be con¬ 
firmed by very general patterns in the language, and not infre¬ 
quently can be firmly established in no other way. A language is 
not an assemblage of unconnected patterns, but a system which is 
integrated to a high degree. The individual patterns will not 
emerge until a number of lesser patterns have been discovered. It 
is, therefore, necessary to make tentative conclusions based on 
relatively slender support, in order to establish a base from which 
they can be confirmed. We must never let the apparent reason¬ 
ableness of such conclusions beguile us into treating them as fully 
established before all relevant data has been examined. 

7.7 As has just been suggested, there are often widespread 
patterns determining the selection of allomorphs of a large number 
of morphemes. It is of course passible to state the facts inde¬ 
pendently for each such morpheme. This may, however, he very 
inefficient, because it requires endless repetition of the same state¬ 
ments. Moreover, it is theoretically unsatisfactory in that it ob¬ 
scures a general pattern. Such general patterns arc important parts 

82 Classing Allomorphs into Morphemes 

of the structure of the language. The objective of a person who is 
analyzing a language is to make such patterns of structure as clear 
as possible. The objective of a person who is learning a language is 
to comprehend these basic patterns in the most thorough way 
possible, and in the shortest time, so that he can manipulate them 
naturally. Both objectives are served by the same type of analysis. 
It is desirable to find simple and general methods of describing the 
distribution of allomorphs. 

One way to make simple statements of variations within a series 
of morphemes is to select one allomorph of each morpheme as a 
base form. Then the other allomorphs can be considered as result¬ 
ing from describable changes from this base form under certain 
statable conditions. Thus, parallel changes in a number of mor¬ 
phemes can be described once. Sometimes it is necessary to append 
a list of the morphemes to which the description of changes applies. 

Changes of this sort are called morphophonemic changes. In 
some languages they arc extensive and complicated, in others rela¬ 
tively few or relatively simple. They are usually referred to as 
“changes,” but this is merely a convenient fiction; they are the 
changes that must be made if you start from the assumed base 
form. They do not represent changes that have occurred in the 
history of the language, unless the base form selected fortuitously 
happens to coincide with an older form. 

7.8 It is sometimes of little importance which allomorph is 
selected as the base form. The English noun plural morpheme 
{-Z,l has three common allomorphs /-z ~ -s ~ -iz/which are pho- 
nologieaily conditioned. Any one of these can be selected as the base 
form. If we assume /-s/ to be basic, we may say that after a voiced 
sound it becomes voiced, /-z/; after /s z § 2 fi J/ a vowel /i/ is 
inserted and, as the vowel is voiced, the morpheme becomes /-iz/. 
Or we could select /-z/ as basic, in which case we would say that 
after a voiceless phoneme (other than /s § 5/) it becomes voiceless 
/-s/. Or we could start with /-iz/ and describe under what condi¬ 
tions the vowel is dropped. One is about as convenient as the other. 

It might be suggested that, since all are equally convenient, we 
should choose that one which is historically correct. If we do this, 
our statements of morphophonemic changes would be also state¬ 
ments of development. Unfortunately, this cannot be done in any 
way that is useful. The reason is that the phonemic system of 

Classing Allomorphs into Morphemes 83 

English has changed;[s] and [z] were at one time not phonemically 
distinct. Any attempt to state the present situation in terms of 
phonemes of Old English, or vice versa, will inevitably falsify the 
facts. Moreover, the historical development of the language, inter¬ 
esting as it may be, is not our present concern. Our purpose is to 
state the existing structure of the language. If we are to turn later 
to historical linguistics, our work then must be based on good 
descriptive statements of the structure of the language at various 
periods. The opposite procedure, basing descriptive work on a his¬ 
torical foundation, is of necessity much less satisfactory and has 
generally worked out to obscure various important details of 

7.9 One of the commonest types of morphophonemic change is 
assimilation. 'Phis is a label for the situation where some phoneme 
is more nearly like its environment than is the phoneme sound in 
the base form. “Similarity” is defined in terms of the phonetic 
description of sounds, as was sketched in 2.12ff and 3.17ff, or in 
terms of a classification of the phonemes directly suited to the 
language concerned. The details of application of such a definition 
will rest on some principles of phonology which will be discussed 
later. The definition can be illustrated by an English word like 
imperfect /imp5rfikt/, which contains a root /porfikt/ and a prefix 
whose base form may be given as (in-). The change of /n/, an 
alveolar nasal, to /m/, a bilabial nasal, makes it more similar to 
/p/, a bilabial stop. The assimilation of /n/ is said to be con¬ 
ditioned by /p/. 

Various relationships exist between assimilated sounds and those 
by which the assimilation is conditioned. As might be expected, the 
two are usually very close together. Most commonly they are 
immediately adjacent in the stream of speech. For example, {in-} 
in intemperate /int^mporit/ : incalcitrant /iijk£lsitr5nt/: impos¬ 
sible /impasibil/ shows assimilation conditioned by the imme¬ 
diately following phoneme. The two commonest allomorphs of 
{-D,} arc an instance of assimilation conditioned by the imme¬ 
diately preceding phoneme, being voiced after voiced sounds as in 
buzzed /b$zd/, and voiceless after voiceless sounds as in wished 
/wiSt/. The two types are sometimes distinguished as progressive, 
in which the assimilated sound follows the conditioning sound, as 
in Turkish gitti ‘he went’ from base form git plus -di; and 

84 Classing Allomorphs into Morphemes 

regressive, in which the assimilated sound precedes the condition¬ 
ing sound, as in Hebrew /mibbayit/ ‘from a house’ from the base 
forms /min/ ‘from’ and /bfiyit/ ‘house.’ 

Assimilation conditioned by an immediately adjacent sound, as 
in Greek /p h leps/ ‘vein’ from /p h leb-/ plus /-s/, is sometimes 
called contiguous in contrast to noncontiguous, in which one or 
more phonemes intervenes between the sounds concerned. The 
latter is much less frequent. A good example is Sanskrit /puspa-ni/ 
'flowers’ where without the assimilation we would expect 
/*puspa-ni/. Under certain circumstances /n/ becomes retroflex 
/n/ conditioned by the occurrence of retroflex /§/ earlier in the 
word. Probably the commonest type of noncontiguous assimilation 
is vowel harmony, in which vowels of successive syllables must be 
similar in some way. This occurs in a limited way in many lan¬ 
guages and in some is very extensive and quite systematic. In 
Hungarian most suffixes have allomorphs exhibiting vowel har¬ 
mony. For example ‘toward’ /-hoz ~ -hez ~ -hoz/ takes the form 
/-hoz/ after all back vowels, /a pnthoz/ ‘toward the shore’; /-hez/ 
after unrounded front vowels, /a kerthez/ ‘ toward the garden ’; and 
/-hoz/ after rounded front vowels, /a folthoz/ ‘toward the earth.’ 

Assimilations also differ in the degree and kind of similarity to 
the conditioning sound. In Arabic the prefix / ? al-/, often trans¬ 
lated ‘the,’ assimilates completely to certain following sounds. 
Thus ‘the peace’ is / ? assala-m/. When the two sounds involved 
differ in more than one feature, assimilation is commonly only 
partial. Any feature or any combination of features may be con¬ 
cerned. English {-Z;} shows assimilation in voice; {in-} shows 
assimilation in point of articulation. The possibilities are best 
shown by an arbitrary example: English /n/ and /p/ differ in three 
respects, alveolar : bilabial, voiced : voiceless, and nasal : stop. 
These may be shown in the form of a three dimensional diagram: 




Classing Allomorphs into Morphemes 85 

/n/ could assimilate to /p/ by becoming: 

m i.e., by becoming bilabial 
d stop 

i? voiceless 

ip voiceless and bilabial 

b stop and bilabial 

t stop and voiceless 

p voiceless, bilabial and stop. 

Of these, the first occurs in English, as we have seen. Two of the 
others are impossible in English because there arc no voiceless nasal 
phonemes such as */$/ and */*P/ in the language. The remaining 
possibilities just do not occur. This is because they are not included 
in the morphophonemic patterns of English, an ultimately arbi¬ 
trary matter, not because they are inherently less easy or less 
natural. Other languages may select other types of assimilations. 

One type of assimilation which is very widespread and occa¬ 
sionally particularly troublesome is palatalization. This is the 
assimilation of velar or dental consonants to high front vowels or 
semivowels, which are similar in articulation to palatal con¬ 
sonants — for example, Italian /mSnaSi/ ‘monks’ from /*m6naki/, 
/16jje/ ‘ he reads ’ from /‘legge/, and /ku6£Ci/ ‘ that ’ from /*ku611i/. 
(/£/ is a palatal lateral.) 

7.10 By no means all of the common morphophonemic changes 
are assimilations. Some are of types so infrequent in occurrence as 
not to warrant special terms; almost any conceivable type of 
change can and does occur. In addition to assimilations, a few 
other kinds are of simple character and can conveniently be given 
generally applicable labels. 

One of these is dissimilation, in which the affected phoneme be¬ 
comes less like the conditioning sound than it might otherwise be. 
For example, in Greek the root for ‘hair’ is /t h rik h -/- In the nom¬ 
inative (the form used as a subject) the suffix /-s/ is added and, 
by a regular morphophonemic change of /*k h s/ to /ks/, this be¬ 
comes /t h riks/. In the genitive the suffix is /-os/. This does not 
condition any change in the /k h /. However, no Greek word ever 
has two aspirate phonemes in successive syllables. The /t h / there¬ 
fore changes to /t/, giving the form /trik b os/ ‘of a hair.’ The 
change of /t h / to /t/ is dissimilation. 

86 Classing AllomorpHs into Morphemes 

Another relatively rare morphophonemic change is metathesis, 
a change in the order of phonemes. For example, there is a prefix 
/hit-/ in the Hebrew system of verbal stem formatives. With roots 
beginning with groove fricatives (/s z § §/), the /t/ of the prefix 
and the fricative change places. This may be seen in /hi§tamm6er/ 
where we would otherwise expect /*hitsamm6er/. 

Much more common than either of these are changes involving 
the addition or loss of phonemes. These are common in many 
languages including English. The /—iz/ allomorph of {—Z x | in dishes 
/di§iz/ may be described as a case of the addition of a phoneme. 
Similar additions are common in Hebrew in such words as /§6bet/ 
‘sitting’ from /*§6bt/. Consonants are commonly dropped in Eng¬ 
lish from what would otherwise be long and awkward (from an 
English point of view!) sequences. For example, lardgrant is com¬ 
monly pronounced /lffengrifent/ rather than /*l&ndgra;nt/ as both 
spelling and etymology would suggest. /gnfcnraotSor/ for grand¬ 
mother is a similar case. In Hebrew certain consonants (/? <; h x r/) 
are not pronounced double even where doubling is morphologi¬ 
cally significant. Thus, /haa’fiS/ ‘the man’ has only one/’/ when 
we might expect two /*ha”ii§/ after the same pattern as seen in 
/hamm&ek/ ‘the king,’ /ha554em/ ‘the name,’ etc. The fact that 
in such cases the preceding vowel is long is sometimes labeled 
“compensatory lengthening." The term should not be so inter¬ 
preted as to suggest some sort of causal over-simplification, but 
merely (as is true of all other morphophonemic changes) as a label 
for certain observed phenomena. Phonemes are not dropped only 
from clusters. In Greek, for example, only a very few consonants 
ever occur at the ends of words. Many which might be expected 
are regularly dropped. Thus the word /stoma/ ‘mouth’ is from 
the root /stomat-/, the formation involving the dropping of the 
final /t/. 

The voicing of consonants between vowels and the unvoicing of 
final consonants are both common. In German the latter is regular 
with all stops. Thus /tak/ ‘day’ is from the root /tag-/. Voicing 
of consonants is often conditioned by the position of the stress, 
as is shown by the two competing pronunciations of English exema 
/6ksima/ and /igziymo/. 

7.11 In many instances it is desirable to select base forms of 
morphemes with care. Often one choice will permit of a rela- 

Classing Allomorphs into Morphemes 87 

tively simple and convenient description of morphology and 
morphophonemics, whereas another will necessitate a much more 
complex and awkward statement. This may be illustrated by the 
following data from Latin: 

/re-ks/ ‘king' /regia/ 'of a king' stem/rc-k-~ re-g-/ 
/greks/ ‘flock' /gregis/ ‘of a flock’ /grek-~ greg-/ 

/noks/ ‘night’ /noktis/ ‘of a night’ /nok- ~ nokt-/ 

/duks/ ‘leader’ /dukis/ ‘of a leader’ /duk-/ 

The suffixes are /-s/ in the forms in the first column, and /-is/ 
in the forms in the second. The stems, except /duk-/ each show 
two allomorphs. These are listed in the third column. The problem 
under consideration is the selection of base forms in such a way 
as to permit the simplest description. The allomorphs show two 
kinds of differences, /k/ : /g/ and single consonant: cluster. 

It would be possible to select the base form independently for 
each stem; but to do so would introduce another complexity. It is 
preferable to select for each stem either the form occurring in the 
first column or that occurring in the second. This will enable us 
to make parallel statements about each. If that in the first column 
is selected, it will prove necessary to state individually for each 
stem what, if any, of these changes apply. On the other hand, if the 
forms in the second column are selected, then the other allomorph 
can in every instance be predicted by a general rule as follows: 
all stem-final consonants become voiceless, if not already voiceless; 
all final clusters are reduced to a single consonant by dropping the 

Given only four sets of forms, as in the data above, the rule 
may seem more complex than a simple listing of the allomorphs. 
However, each of these is representative of a number of others. 
Given the whole list of similar words in the language, the rules are 
far simpler. 

7.12 As a general rule of thumb, an analysis should be based 
on that series of forms which shows the greatest diversity. In the 
case just described, that is the second column. These words exhibit 
the following endings of the stems: /gk kt/, whereas all forms in 
the first column have stems ending in /k/. It will ordinarily be 
easier to state how diversity replaces uniformity than to describe 
the ways in which such differences are suppressed. 

88 Classing Allomorphs into Morphemes 

The same rule of thumb is also useful in language learning. 
In learning vocabulary comparable to that in the example above, 
learn that set of forms which shows the greatest diversity. It will 
usually be easier to associate the other forms of each word with 
this base form than with any other. That is, /re-gis/ carries all the 
structural information which /re-ks/ does, plus an indication of 
the morphophonemic changes in the stem which arc not clearly 
shown in /re-ks/. If /re-ks/ is learned, you must learn in addition 
what changes must be applied to produce /re-gis/. 

In some instances there is no single set of allomorphs from which 
all the others can be predicted in this way. This is true for some 
other Latin nouns not included in the data above. When this occurs 
it is necessary to learn either more than one form for each stem, or 
one form plus some indication of the changes to be expected. 
Common practice in many languages is the former. This is the 
reason for learning principal parts of verbs in German, Latin, or 
Greek — that is, a selection of those forms of each verb which 
indicate the necessary morphologic and morphophonemic facts 
about each. 

7.13 In the last several paragraphs something was said about 
“rules.” The term is open to misinterpretation to such an extent 
that many linguists studiously avoid using it at all. The trouble 
arises from the fact that many people have expected “grammar” 
to tell them how a language “ought” to be spoken. Of course, this 
is no proper function of descriptive linguistics. If it can be clearly 
understood that a “rule” is merely a statement of what does occur, 
without any connotation of what should occur, the term is harm¬ 
less. A legislative type of grammar is of very questionable value to 
a person who speaks the language, and absolutely worthless to 
a learner. 

7.14 In 7.3 and the sections following it a criterion for classing 
allomorphs into morphemes was presented; however, its applica¬ 
tion was specifically restricted to those instances in which the con¬ 
ditioning factor is phonologic. When the conditioning is mor¬ 
phologic, different criteria must be applied. Two elements can be 
considered as allomorphs of the same morpheme if: (1) they have 
a common meaning, (2) they are in complementary distribution, 
and (3) they occur in parallel formations. Note that there are three 
requirements. All three must be met. The third requirement in 

Classing Allomorphs into Morphemes 89 

particular is inadequately stated, but a better statement would 
involve a number of complexities. Some of these, and some of the 
other possibilities in morphologic conditioning, are beyond the 
scope of an introductory treatment. 

7.15 The meaning of this criterion is best shown by some ex¬ 
amples. The three allomorphs of the English plural J-z ~ —S -iz/ 
can be established as members of the same morpheme by the 
criterion presented in 7.3, since they are phonologically condi¬ 
tioned. /-in/ is another allomorph which can be grouped with them 
for the following reasons: (1) The meanings are alike, ‘plural.’ 
(2) They are in complementary distribution — /-in/ occurs only 
with ox; none of the others do. (3) The constructions in which they 
occur are parallel; /-in/ occurs only in the construction oxen 
/aksin/, while cows /kawz/ is representative of the constructions 
in which /-z ~ -s ~ -iz/ occurs. These can be said to be parallel 
because /&ks/ and /kuw/ occur in some identical and many similar 
environments, and so likewise do /aksin/ and /kawz/. This sug¬ 
gests that /-in/ and /-z/ have the same functional place in the 
grammatical structure of the language. 

7.16 As a second example we may take the problem raised 
in 6.13. Two affixes, /-huu/ and /-6o/, both meaning ‘him’ were 
found. They were there assumed to be allomorphs of the same 
morpheme. The justification for this assumption depends on the 
criterion just presented. (1) Their meanings seem to be alike. 
(2) They are in complementary distribution: /-6o/ occurs after 0 
‘he’ and /—t—/ ‘thou’ (in /zakartdo/ ‘thou remembered him’), 
whereas /-huu/ occurs after all other suffixes expressing the actor. 
Because this distribution is morphologically conditioned, it is 
necessary to consider the third requirement. (3) They occur in 
parallel constructions. This may be seen most clearly if the mor¬ 

phemes are separated thus: 
/z-k-r/ 4- /-o-aa-/ + 

/-uu/ + /-huu/ /zokaaniuhuu/ 
0 + /-do/ /zokaardo/ 

‘they remembered him’ 
‘he remembered him’ 

/z-k-r/ + /-o-a-/ + 

/—tli/ + /-huu/ /zokartfihuu/ 
/—t—/ + /-do/ /zokartdo/ 

‘I remembered him’ 

‘thou remembered him’ 

90 Classing Allomorphs into Morphemes 

These four forms show a parallelism in internal construction. In 
addition they occur in very similar environments so that they also 
show the type of parallelism which was found in the case of /aksiu/ 
and /k&wz/. 

7.17 A third example will also be drawn from Hebrew: 

/zaakartii/ ‘I remembered' /lii/ ‘I have' 

/zaakartaa/ ‘thou remembered’ /lokaa/ ‘thou hast’ 

From these (preferably of course by comparison with additional 
data) we can isolate two elements meaning ‘I,’ /—tii/ and /—ii/, 
and two meaning ‘thou,’ /-taa/ and /-okaa/. In addition to the 
similarity of meaning they seem to be in complementary distribu¬ 
tion, though of course it is exceedingly rash to base any such con¬ 
clusion on such slender evidence. They are not, however, in parallel 
formations — /l—/ + /-ii/ is not parallel with /z-k-r/ 4- /-aa-a-/ 
+ /-tii/. There is nothing to support considering /-tii/ and /-ii/ 
as members of the same morpheme, nor /-taa/ and /-okaa/. Note 
that the conclusion is stated negatively. It may be that further 
investigation will show that, while these particular constructions 
are not parallel, they do actually occur in comparable structures. 
This does not happen to be the case, but the data at hand does not 
prove that it is not. 

7.18 Some morphemes are highly various, while others have 
only a single allomorph in all environments. Moreover, allomorphs 
may be quite similar or wholly unlike. Both conditions can be 
illustrated in English. The prefix in intolerant, impossible, in¬ 
coherent, illegal, etc., has the forms /in- ~ im- ~ iQ- ~ i-/. By 
contrast the morpheme {-Zi} ‘plural’ includes such diverse allo¬ 
morphs as those in cows, oxen, stomata, mice, and deer, /-z ~ -in 
~ -to ~ ay <— (aw) ~ 0/. 

7.19 A special case of affixes of highly variable form, reduplica¬ 
tions in which the affix is identical with some part of its environ¬ 
ment, can be seen in the Tagalog words in the following list: 

/isa/ ‘one’ /iisa/ ‘only one’ prefix /i—/ 

/dalawa/ ‘two’ /dadalawa/ ‘only two’ /da-/ 

/tatl6/ ‘three’ /tatatld/ ‘only three’ /ta-/ 

/piso/ ‘peso’ /pipiso/ ‘only one peso’ /pi-/ etc. 

Classing Allomorphs into Morphemes 91 

Reduplications can be prefixes, infixes, or suffixes. An example 
of the latter can be seen in the following Mam words. In all the 
words cited there is a suffix /-r)/ added after the reduplication. 

/sp’iliq/ ‘it is smooth’ /sp’ililiij/ ‘it is slippery’ suffix /—li—/ 

/toloi]/ ‘roll (a log)’ /tololoij/ ‘water is rolling’ /-lo-/ 

/tunuij/‘hang’ /tununuij/ ‘it went out hang- /-nu-/ 


In the examples of reduplications, several different words were 
cited. If this is not done, affixes may sometimes appear to be 
reduplications by accident. For example the English word singing 
/slijii]/ from /slij/ could, if nothing more were known, be inter¬ 
preted as a reduplication. But if other words are compared — per¬ 
haps riding /raydiij/, going /gdwig/ — this interpretation is easily 
seen to be untenable. The significant thing about a reduplication 
is the allomorphic variation which occurs; reduplications can only 
be identified when sufficient data is available to indicate the extent 
and nature of such variation. 



0<5<X>00<X><> < £0<>0<><>00<><X>0<>0 


Outline of English 

8.1 English grammar is traditionally described in terms of eight 
parts of speech: nouns, pronouns, adjectives, verbs, adverbs, prepo¬ 
sitions, conjunctions, and interjections. These eight classes are of 
quite diverse character and validity. The familiar definitions over¬ 
lap and conflict, or are so vague as to be nearly inapplicable. Some 
parts of speech gather together a number of not very obviously 
related types of words. In other cases, the line of demarcation be¬ 
tween parts of speech is rather arbitrary. Before any thorough¬ 
going description of English morphology can be attempted, it is 
necessary to refine these definitions, recognize the differences in 
status between the classes, and make various other alterations. 

8.2 One of the sources of difficulty with the English parts of 
speech is the tradition of defining them on the basis of meaning. 
Thus, nouns are defined as words that “name persons, places, or 
things." There are numerous nouns that have no such reference: 
goodness, home run, fatherhood, etc. To cover these, the definition is 
commonly expanded by the addition of some such list as “. . . 
qualities, action, relationships, etc.” The “etc.,” of course, imme¬ 
diately disqualifies this description as a definition, but this criticism 
is trivial. The serious problem is that adjectives are defined as 


Outline of English Morphology 93 

words that “indicate a quality,” and verbs as words that “specify 
actions, states, feelings, etc.” Since “quality” and “action” each 
occur in two of these definitions, each fails as a basis of discrimina¬ 
tion between parts of speech. Presumably a word referring to a 
“quality” cannot be a verb, unless the “etc.” will cover such a 
case. But whether a word is a noun or an adjective cannot be 
determined from the reference to a “quality.” If such differences 
are ruled out as nondistinctive, recourse must be had to an ap¬ 
parent distinction between “name,” “indicate,” and “specify.” 
These are the only distinctive elements in the three definitions. 
Yet there is no evidence that these words are used with sufficient 
precision to warrant resting any significance on the difference, if 
any such is in fact discernible. 

In short, the traditional definitions of parts of speech are largely 
unworkable. In practice, these overlappings and confusions have 
little effect. Ultimate recourse is always had to other criteria. The 
assignment of tallness to a certain part of speech docs not rest on 
whether it “names” or “indicates” a “quality,” but upon its 
linguistic behavior. In short, tallness is used as a noun. Every 
American knows this, and hence assumes that it must “name” 
rather than “indicate” (or whatever distinction is made in the 
particular set of definitions which he is attempting to apply). Con¬ 
versely, tall is considered to “indicate” rather than “name” be¬ 
cause it is used as an adjective. The controlling criterion is always 
usage, not meaning, a fact which is being increasingly widely 
recognized among scholars of the English language. 

8.3 The second difficulty with such definitions is more serious. 
Our one reason for desiring to classify words into parts of speech is 
to simplify our description of the structure of the language. To 
achieve this simplification we must class together words which play 
essentially identical rdles in the structure of the language. If this 
is done, statements can be made about the structural relationships 
of a class as a whole which will apply rather precisely to all mem¬ 
bers individually. But if they do not have structural function in 
common, then the classes become useless for our purpose. In fact, 
the usefulness of a classification depends directly on the uni¬ 
formity of structural relationships of the words included. Words 
of similar meaning do not necessarily have common function. Con¬ 
sider tall and tallness . These certainly both refer in some vague and 

94 Outline of English Morphology 

indefinable way to a quality. Suppose we were to set up a class of 
“quality words” including these and many others. There would 
be nothing useful we could do with such a class, at least not in 
morphologic analysis. Some (e.g., tall) would occur in such an en¬ 
vironment as The - man came.; others (e.g., tallness ) in- is 

ta be desired. Only very few environments could be suggested in 
which all could occur, and even fewer, perhaps none, in which 
“quality words” could occur but not members of other classes. 
In other words, tall and tallness have very little in common gram¬ 
matically which is distinctive of any class including them. 

8.4 The only classes that will necessarily be grammatically 
useful are those which are grammatically defined. Others may, by 
more or less fortuitous coincidence, not be obviously bad. This is 
unfortunate because it can be so misleading. Most nouns (gram¬ 
matical!}' defined) do in fact refer to “persons, places, or things.” 
This is a useful fact to know, and it is not wholly accidental. Never¬ 
theless, nouns are not nouns because of this, but because of some¬ 
thing else which is only rather distantly connected, namely the 
grammatical structure of English. The meaning-based class and 
the grammar-based class are different in just enough cases to cause 
endless confusion if the two are not clearly distinguished in any 
discussion of the language. 

8.6 At this point a quite legitimate objection can well be 
raised: is not our chief interest in language its use as a means of 
conveying meanings? Certainly, one very easily justifiable reason 
for language study is to understand it in this capacity. Why then 
do we set meaning aside so brusquely? 

We must answer these questions by saying that we rule meaning 
out of our analysis precisely because it is meaning in which we arc 
ultimately interested. We hope that the study of the structure of 
English, or of any other language, will lead ultimately to an under¬ 
standing of the meanings it conveys, as far as they may be ac¬ 
cessible to us. We also hope that it will make them more accessible 
to us than they were before such study was undertaken. But, if our 
understanding of structure is based on meaning, and our under¬ 
standing of meaning on structure, wo are in a most vicious circle. 
All kinds of meanings can be read out, if in the course of analysis 
they are first put in. In order to make the study of meaning as 
effective as possible, we must first have an objective understanding 

Outline of English Morphology 95 

of structure. Only a person wholly uninterested in any ultimate 
understanding of meaning can afford to base his structural analysis 
on meaning to any greater extent than may be unavoidable. 

8.6 Returning to English parts of speech, we find that an 
attempt to find definitions based on grammar necessarily resorts 
to two types of criteria. Some words (we will continue for the 
present to use word in the rather loose familiar sense without close 
definition) are found in a small set of related forms. These sets fall 
into four patterns, each of which is a paradigm. They may be 
typified by the following examples: 

man /ms6n/ I /ky/ ride /r-iyd/ fine /ikyn/ 

men /m6n/ me /miy/ rides /raydz/ finer /faynor/ 

man's /minnz/ my /may/ rode /r6wd/ finest /f&ynist/ 

men’s /mdnz/ mine /mayn/ ridden /rfdin/ 

riding /raydirj/ 

The first of these is representative of a large class of words, in 
each of which occur four forms corresponding to the four forms of 
man, for example: 

ox /aks/ child /fiayld/ boy /boy/ table /tdybil/ 

oxen /&ksin/ children /fildrin/ boys /b5yz/ tables /tdybilz/ 

ox’s /&ksiz/ child’s /fiayldz/ boy’s /b5yz/ table's /tdybilz/ 

ozoa's /fiksinz/ children’s /bildrinz/ boys’ /boyz/ tabics’ /tdybilz/ 

Any of the thousands of words which occur in this paradigm may 
be defined as a noun. Similarly, there are a few others like I, and 
these may be defined as personal pronouns. A very large number 
occur in a paradigm like that of ride and may be labeled verbs. 
A moderate number are like fine and are called adjectives. Note 
that by this definition beautifid is not an adjective, because it docs 
not occur in the paradigm beautiful, *be.autifuler, *beautifulest. At 
least, these forms are not heard in the speech of most Americans. 
For the few who use them, of course, beautiful is an adjective. 

These four paradigms define four paradigmatic classes. The 
remaining parts of speech, so far as they have any validity, are of 
a dilferent sort. They are classes of words which occur in the same 
or comparable environments in English utterances. They arc 
syntactic classes. Syntactic classes sometimes comprehend para¬ 
digmatic classes. For example, we have just pointed out that 

96 Outline of English Morphology 

beautiful is not in most dialects an adjective. It does, however, 
occur in the type of environments in which adjectives are found. 
It therefore belongs to a large syntactic class which also includes 
the adjectives. This class we will call adjectivals. Some features of 
syntactic classes will be mentioned in Chapter 10. 

8.7 We may abstract from each paradigm a set of morphemes 
by which it is formed. Leaving the pronouns with their complica¬ 
tions aside for a moment, the paradigms may be summarized as 

noun stem verb stem 

noun stem {-Zi} verb stem {-Z 3 } 

noun stem {-Z 2 | verb stem {—Di} 

noun stem {-Z,Z 2 | verb stem {-D 2 } 

verb stem {—ig} 

These affixes are inflectional suffixes. They may be contrasted 
with derivational affixes, a term which covers all other suffixes 
and all prefixes in English. They serve to form stems that can 
function in these paradigms, or to form words of other classes. 
For example /-ayz/ is a verb-stem-forming derivational suffix. 
Every word containing /-ayz/ is a verb and can be found in the 
five forms of the very paradigm, unless some further derivational 
suffix has been added. For example, phonemicize is a verb; phonemi- 
cizalion with an additional suffix, the noun-stem-forming /-eysin/, 
is a noun. 

English inflectional suffixes typically do not form stems. That 
is, once such a suffix is added, no other prefix or suffix (except 
j-Zi) after {-Zi|) can be added. Thus rereads is a member of a 
paradigm reread, rereads, reread, reread, rereading based on the 
verb stem /riyriyd/, not a formation from reads by the addition of 
the prefix /riy-/. Inflectional suffixes are an outer layer in word 

8.8 This suggests three convenient divisions of English gram¬ 
mar. The first will be the analysis of the paradigms of the four parts 
of speech: noun, pronoun, verb, and adjective. It will involve 
largely the discussion of the inflectional suffixes and their al- 
lomorphs. The second will be the analysis of the formation of 
stems and of words which are not inflected. In this short outline 
very little can be said about this division. These two together 

adj. stem 
adj. stem {-or} 
adj. stem {-ist} 

Outline of English Morphology 97 

comprise morphology. The third will be the description of the 
longer sequences into which the words enter. This is called syntax 
and will form the subject of Chapter 10. 

Of course, these three are not wholly distinct nor always clearly 
separable. But in general they constitute the most useful divisions 
of the subject matter of English grammar. In other languages 
similar divisions may be found, or one or the other of these may 
be missing, or there may be other divisions. The utility of such an 
organization of the subject depends on the structure of the lan¬ 
guage. Such a division is practically dictated by English structure, 
but may be worse than useless in another form of speech. 

8.9 As noted above, English nouns have two inflectional 
suffixes, j-Z]} and {-Zj}. These two have, however, very different 
standing. There is a large number of words which can be found 
only with {-Zi} and never with j-Z 2 j. Many more can be found 
much more commonly with j-Z,} than with j-Zo}. There are only 
a few words with which j —Z 2 J is at all common. Because of this, 
it will prove much more useful if the class of nouns is defined as all 
those words which occur in the first half of the paradigm, that is, 
those which have both singular and plural. After so defining the 
class of nouns, it must be noted that many occur in two more 
forms involving the affix {-Z 2 1. 

The reason for the comparative rarity of j-Z 2 | is quite obvious. 
This suffix is in competition with of, and the trend seems to be 
toward increased use of of and consequently increasing restrictions 
on the use of {-Z 2 |. Many noun stems are now used only with of; 
in many others, of is far commoner than |-Z 2 }. Moreover, there 
are considerable variations of usage from dialect to dialect and 
from speaker to speaker. This is more or less inevitable when usage 
is undergoing relatively rapid change. English seems in fact to be 
in the process of losing j-Z*}, as it has already lost a large number 
of other inflectional affixes during the last millennium. 

8.10 The plural affix of the noun, j-Z.}, has a very large num¬ 
ber of allomorphs. Instead of a detailed listing they will be de¬ 
scribed in general groups. 

1. /-z ~ -s ~ -iz/ These three are by far the commonest, occur¬ 
ring with the overwhelming majority of noun stems, including 
most of the new words which are produced each year. Though 
many of the stems using other allomorphs are among the com- 

98 Outline of English Morphology 

monest nouns in the language, the large majority of plurals in 
any body of English will be formed with members of this group 
of allomorphs. Within the group they are phonologically con¬ 

/-z/ occurs after stems ending /b d g v c5 m n ij r 1 o y w h/ 
/kbb kob z/ cub cubs /biy bfyz/ bee bees 
/-s/ occurs after stems ending /p t k f 0/ 

/k5p k5ps/ cup cups /kl6f kl4fs/ clef clefs 
/—iz/ occurs after stems ending /s z s l 5 ]/ 

/gli§s glifcsiz/ glass glasses /\vi6 with/ witch witches 

2. /-z ~ -iz/ plus a change of the final consonant of the stems. 

/z <— (s)/ in one word only 
/haws hawziz/ house houses 
/v «— (0/ in about a dozen words 
/nfiyf nfiyvz/ knife knives 
fS <— ( 0 )/ in about eight words 
/pseO p;eSz/ path paths 

3. /-in/ with or without additional changes in three words. 

/aks aksin/ ox oxen 
A&yld Cildrin/ child children 

/br5Sor br<S5rin/ brother brethren (Only in specialized senses; 
as a common kinship term the plural /br6<5orz/ brothers is 

4. Various replacives in a few very common nouns. 

/e <-(*)/ 

/iy«- (u)/ 

/iy <— (uw)/ 
/ay <- (aw)/ 

/man mdn/ ma?i men 
/wumin wimin/ woman women 
/fut fJyt/ foot feet 
/giiws giys/ poosc geese 
/maws mays/ mouse mice 

o. Zero in a few nouns, mostly referring to animals. Some of these 
are pluralized in this way by some speakers and with /-z ~ -s 
~ -iz/ by others; and some speakers use both forms in different 

/Slyp siyp/ sheep sheep 

Outline of English Morphology 99 

G. Certain loan words from other languages, mostly Latin, have 
retained the plural formation used in the original language, at 
least in the spelling. There is a strong tendency to make these 
conform to the English pattern by changing the form of 
{-Zi} to /-z ~ -s ~ -iz/. 

From Latin: /ohSmno olomniy/ alumna alumnae 
/kifektos ksfckt&y/ cactus cacti 

/kruysis kraysiyz/ crisis crises 

/Ind&ks Indislyz/ index indices 

/diktom dlkto/ dictum dicta 

From Greek: /stbwmo stdwmoto/ stoma stomata 
/kritlriyon kritfriyo/ criterion criteria 
From Hebrew: /6<5rob edrobim/ cherub cherultim 

8.11 The listing of the noun plural formations just given does 
not include a full list of the allomorphs of the inflectional morpheme 
|-Z,}. It does, however, suggest some of the items that would have 
to be included in it. The significant thing is the nature of the 
variation that is shown. The first group is by all odds the com¬ 
monest. It defines a subclass of nouns into which most new words 
find their way. In addition, this type of plural formation occurs 
as an alternative to any of the others, whenever there is any varia¬ 
tion in usage. Such a subclass, defined by an allomorph or a 
phonologically conditioned set of allomorplis, may be called a 
productive subclass. 

Groups 2 to 5 in the tabulation represent vestiges of older 
English patterns which were once much more widespread than they 
now are. In the course of several centuries, large numbers of nouns 
which were formerly pluralized in ways comparable to those listed 
in groups 2 to 5 have altered their paradigms and come to use the 
/-z ~ -s ~ -iz/ type of plural. For example, the Old English form 
which became modern cow had a plural which, if it had continued 
to modern times, would be expected to be pronounced /*kay/. 
That is, cow would follow the same paradigm as mouse. Instead, at 
some point in the history of the language, this plural was replaced 
by that which developed into /kawz/ cows. Such a process of alter¬ 
ation is likely to affect the less common words first. As a result, a 
few, but mostly very common, words continue to use the less 
common allomorphs. This is quite typical of the situation to be 

100 Outline of English Morphology 

found in many languages. The “irregular” forms (those belonging 
to smaller paradigmatic subclasses) are usually the commoner 
words of the language. 

Group 6 is a somewhat different case. These are recent loan 
words from classical languages. They are mostly technical or 
learned words, used only by people with classical education or in 
contact with classical traditions. Because of this, they succeed par¬ 
tially in retaining unassimilated paradigms in spite of their com¬ 
parative rarity. The few which have become part of the common 
vocabulary of the American public have almost all received new 
plurals. Sometimes the Latin singular has become the stem, some¬ 
times the Latin plural. Many people use criteria as a singular from 
which a plural /kr^ytfriyoz/ is formed (the pronunciation varies 
widely!). Data is very commonly used as a singular, from which 
the plural /d£toz/ can be formed. Indeed, many people use data 
and datum side by side in different meanings without recognizing 
the relationship between them. Only the eternal vigilance of classics- 
minded teachers, editors, and scholars has preserved these irregular 
(for English) plurals, and even that has been powerless in a large 
and increasing number of cases. 

The process involved is illustrated in the case of the word shmoo. 
A1 Capp, in coining and publicizing this word through his comic 
strip, stated clearly and repeatedly that the plural was to be 
shmoon, presumably /smuwn/. The effort was entirely unsuc¬ 
cessful. Shmoo was readily accepted, shmoon not at all. The Amer¬ 
ican readers immediately formed the plural /smtiwz/, thus assign¬ 
ing it, as they felt new words should be, to the productive subclass. 
Formation of new words is a ceaseless activity in English. Anyone 
has both a right to indulge in it and a reasonable prospect of 
success. Tampering with the morphologic patterns of the language 
is probably equally within an American’s constitutional rights, 
but the chances of successful promulgation are infinitesimal. 
Morphology is much more basic to language than is vocabulary. 

8.12 The second noun inflectional suffix {-Z 2 | has only four 
allomorphs. /-z~-s~-iz/ are phonologically conditioned and 
follow the same distribution as the similar allomorphs {-Zi] and 
{—Za}: Rosa’s /r6wzas/, Jack's /jifeks/, Rose’s /r6wziz/. 0 is used 
after {-Zi} when the latter ends in /z/ or /s/, but not after other 
allomorphs: boy's /b6yz/, cats /ks§ts/, men's /m6nz/. In addition, 

Outline of English Morphology 101 

the 0 allomorph is occasionally used after proper names ending in 
/z/ or /s/: James’ /]6ymz/ or /feymziz/. With the majority of 
noun stems {-Z,}, [-7*], and {-Z.Zjj will be exactly alike. That 
is, the full noun paradigm is commonly similar to boy boys boy's 
boys’ /b 6y b5yz b5yz b$yz /. (The spelling distinction is, of course, 
purely arbitrary.) The non-distiuctive form of {-7n\ in so many 
instances is perhaps a contributory factor in its declining use. 

8.13 The verb paradigm presents some similar problems, as 
well as some that are rather different. Of the four inflectional 
affixes, {-iij \ is unique in having only one allomorph in most forms 
of English, but varying widely from one dialect to another. {-Z 3 | 
has only three /-z ~ -s ~ -iz/. These are phonologically condi¬ 
tioned throughout, and hence require little discussion. In three 
verbs, however, there are minor irregularities, do docs /duw d5z/, 
have has /hi£v hrez/, and say says /s6y sdz/. We may consider these 
cases as formed by three special allomorphs of |-Z 3 }, /z/ plus /a 

(uw)/| /z/ plus /0 «— (v)/, and /z/ plus /e <— (ey)/; or we may 
consider the changes as part of the stem, in which case each stem 
has two allomorphs /duw ~ do-/) /htev ~ hie-/, and /sey ~ se-/. 
The latter has much to commend it, since the same stem forms 
occur before {-D 2 } in done /d5n/, had /hied/, and said /s6d/. 
Either analysis is acceptable. In the next section, however, all 
changes will be considered as assignable to the affix. 

8.14 Both |-Di} and |-D»} have numerous allomorphs of 
various types. In the majority of instances the two forms are 
exactly alike, but in a few verbs they are different, and so must be 
considered as two morphemes. Both are like {—Z x | in having a set 
of phonologically conditioned allomorphs /-d ~ -t ~ -id/ which 
taken together define the productive subclass of verbs. The re¬ 
maining allomorphs are all morphologically conditioned, and all 
have rather restricted distributions. In the tabulation below, some 
statistics on occurrence will be given. These are based on the 
author’s personal usage. The figures are, of course, not necessarily 
the same as might be found in another person’s speech. There are 
three reasons for possible differences: (1) Some verbs commonly 
listed as “irregular” have been omitted because I do not regularly 
use them. Another person might have a different active vocabulary. 
(2) Some verbs have two paradigms, both acceptable, and com¬ 
monly others which are considered as substandard. Another person 

102 Outline of English Morphology 

might select differently among the alternatives. (3) Since the 
analysis is based on phonemic transcription, dialectal differences 
in pronunciation will affect it in many, though usually minor, ways. 

The verbs of English may be classified into the following sub¬ 
classes based on differences in paradigms. These arc paradigmatic 
subclasses. They are arranged in order of descending size. 

1. Di} and {-D 2 j = /-d~-t~-id/ with the following dis¬ 

/—d/ after /bgjvSzimnijlray wh/ 

/r£b r5bd rSbd/ rub nibbed rubbed 

/-t/after /pkSfOsS/ 

/st6p st^pt stdpt/ step stepped stepped 
/-id/ after /t d/ 

/sfyt slytid slytid/ seat seated seated 

2. {-Djj and j-D 2 } = 0 in nineteen verbs: 

bet burst cast cost cut hit hurt let pul quit rid set shed shut 
spit split spread thrust wet 
/k$t k$t k$t/ cut ad cut 

3. [-DiJ and {-D 2 j = /o <— (i)/ in fourteen verbs: 

cling dig fling shrink sink (transitive) sling slink spin sti?ig 
stink string swing win wring 
/spin sp5n spSn/ spin spun spun 

4. {-Di} and j-D 2 } = /—t/ plus /e<— (iy)/ in nine verbs: 

creep deal feel keep leap mean sleep sweep weep 
/mlyn ntent ntent/ mean meant meant 

5. {-D,} and {-D 2 } = /e <— (iy)/ in eight verbs: 

bleed breed feed lead meet plead read speed 
/11yd ted ted/ lead led led 

6. {-Di j - /ffi <— (i)/ and {-D 2 J ■= /o«— (i)/ in seven verbs: 

begin drink ring sing sink (intransitive) spring swim 
/drlijk drteqk dr5qk/ drink drank drunk 

7. {-D,} = /ow <— (ay)/ and |-D 2 } /-in/ plus /i<-(ay)/ in 
seven verbs: 

drive ride rise smite strive thrive write 
/rayd r6wd rldin/ ride rode ridden 




Outline of English Morphology 103 

8. {-Dij and j-D 2 j = /t <— (d)/ in six verbs: 

bend build lend rend send spend 
/s6nd s6nt s6nt/ send sent sent 

9. {-Di} = /ow<— (iy)/ and {-D.} - /-in/ plus /ow <— (iy)/ in 
four verbs: 

freeze speak steal weave 

/spiyk sp6\vk sp6wkin/ speak spoke spoken 

10. {-Di} and {-D 2 } = /aw ♦— (ay)/ in four verbs: 

bind find grind wind 

/b&ynd bawnd bawnd/ bind bound bound 

11. {-Di} =* /uw <— (ow)/ and {-D 2 | = /-»/ in four verbs: 

blow grow Icnow throw 

/n6w nuw n6wn/ know knew known 

12. |-Dij = /oh *— (e)/ and JD 2 } = /-n/ plus /oh ♦- (e)/ in four 

bear swear tear wear 

/X&r t6nr t5iirn/ tear tore torn 

13. {-D,} = /u <- (ey)/ and {-I) 2 j - /-in/ in three verbs: 

forsake shake take 

/t6yk tuk t<5ykin/ take took taken 

14-19. Six subclasses, each containing two verl>s. 

20-53. Thirty-four subclasses, each containing only a single verb. 

8.15 One of the latter single-membered subclasses comprises 
the verb be. This is exceedingly irregular, and in addition has cer¬ 
tain additional forms which are not distinguished in other verbs. 
The forms may be listed as follows, using ride as a comparison: 


/biy ur aim/ 

be are am 

— I-Z,} 






/w5z w5r/ 

was were 

-{-D s | 




— l-hal 




104 Outline of English Morphology 

The verb be is exceedingly common and highly specialized in its 
usages. These two facts combine to make possible continued re¬ 
sistance against conformity of be with the prevailing patterns. 
Such additional complications are quite characteristic of such 
highly specialized and common words in all languages. 

8.16 There is a small group of words, can could will would shall 
should may might must , which are traditionally included with the 
verbs. By the definition used here, it is impossible to classify them 
as verbs since they show none of the verbal inflection, with the 
possible exception of {-Di}. That is, some people consider could 
as can plus {—IDi}, and similarly, would should might as the past 
forms of will shall may. There is doubtful value in this analysis, 
but in any case the class is quite distinct from verbs in many other 
respects and quite uniform within itself in usage, and so must be 
recognized as a clearly marked class in English structure. Whether 
it is treated as a highly specialized subclass of verbs (auxiliary 
verbs) or as a separate class closely associated with verbs (verbal 
auxiliaries) docs not matter greatly. We will here elect the latter 
alternative. The definition of a verbal auxiliary must be based 
largely on syntax rather than on the somewhat debatable inflection, 
and is therefore a syntactic rather than a paradigmatic class, 

8.17 The inflection of adjectives is relatively simple and regu¬ 
lar, {-or} and {—ist} being constant in form in the vast majority of 
adjectives. Only with a few are there any irregularities. In these 
cases there is variation not only in the form of the suffix, but also 
in the form of the stem. Only one requires attention here: 

/gud b6tor bfet/ good belter best 

Probably the best way to analyze this is to assume three allomorphs 
of the stem /gud ~ bet-~ be-/. The suffix {-or} is then of the 
usual form, while {-ist} occurs in the allomorph /-st/. The sort of 
complete change in form seen in the stem has been called supple- 
tion. The term, as such, is not particularly important, but the 
phenomenon does deserve some attention. It can be a source of un¬ 
necessary confusion to students. There is, after all, no reason 
whatever why a stem cannot have widely divergent allomorphs. 
Considering the languages of the world, we find the phenomenon 
very widespread, though seldom very common in any particular 
language. The English verb go went gone /g6w w6nt g$Hn/ is 

Outline of English Morphology 105 

another familiar example. The verb fero luli laius ‘carry’ is one of 
the plagues of Latin students, as is /p h ero- oisa lue-neka/ ‘carry’ 
for Greek students. 

8.18 The personal pronouns are eight in number. Each of them 
occurs in a paradigm of four forms. While it is, of course, possible 
to analyze these into stems and affixes, the small number of items 
and the high degree of complexity makes such a procedure some¬ 
what questionable as a practical device. The paradigms are as 

































































These are subject to considerable variation from speaker to 
speaker. Such forms as /yoHrn hizin h5rn/ sometimes replace 
/yonrz hiz h5rz/ as the fourth form, but since they arc considered 
quite inelegant, they are usually avoided. While the older distinc¬ 
tion between who and whom is frequently preserved in writing, the 
distinction /huw/ : /htiwm/ is very seldom maintained in speech. 
Some people use /huwm/ after prepositions but not in other places 
where /miy/ or /him/ would be used. Moreover, /huwm/ fre¬ 
quently carries some taint of pedantry or snobbishness, so that 
some people deliberately avoid it. 

8.19 In addition, most of the personal pronouns vary quite 
widely in the pronunciation of any given speaker, different forms 
being used in different contexts. The pronunciations indicated in 
the tabulation are those which I use when the pronouns are cited 
in isolation. In connected speech the stress is mast commonly /'/ 
or /“/, and the vowels and consonants often differ from thase in 
the base forms listed. Few Americans are aware of the extent of the 
variation. This is partly because the spelling seldom reflects the 
differences. 'Em is occasionally written for thorn, but not nearly 
as frequently as the pronoun is pronounced /om/. This spelling 
gives some suggestion of substandard speech, and any others, for 

106 Outline of English Morphology 

example ’im for him, have a strong implication of this sort. Actually 
/im/ is with most people the commonest pronunciation of him. 
There is no difference between educated and uneducated in this; 
all alike very rarely use pronunciations with /h/. Each item in the 
table must, therefore, be understood to represent a small group 
of pronunciations varying in accordance with complex morpho- 
phonemic patterns. 

8.20 There are a few vestigial remains of other inflections or of 
other inflected words. As typical may be cited the old personal 
pronoun thou thee thy thine /Saw ?Sfy Say Sayn/ now practically 
restricted to use in oral reading of older literature or in prayer. 
With it is associated an additional inflectional suffix used with 
verb stems. This takes the form /—ist/ or /-est/. The pronuncia¬ 
tion varies somewhat, partly at least because of the unfamiliarity 
of most speakers with the forms. 

More conservative members of the Society of Friends have pre¬ 
served this pronoun, but generally in a special “plain speech” 
paradigm /Sfy Siy bay Sayn/. The accompanying verb is used in 
the {-Zjj form. This usage is probably best interpreted as a partial 
assimilation to prevailing patterns, perhaps influenced by the 
similar spread of the old second form of the paradigm /y\y ydw 
your ySnrz/ to replace the old first form. Quaker “plain speech” 
arose from a deliberate attempt to resist the developing patterns 
of the language. Ironically, the paradigm was formed on the 
analogy of the usage they were resisting. 

8.21 The remaining parts of speech must be defined on the 
basis of syntax. This is in many ways a much less sure basis. The 
possible syntactic uses of words are so many, and often so varied, 
that it is quite difficult to determine which are the significant ones 
on which the classification should be based. There is accordingly 
less possibility of reaching a classification on which all will agree, 
though the broad outlines are generally clear. For example, the 
group of words traditionally called prepositions are a reasonably 
definite class with characteristic syntactic uses. They are used 
before nouns and at the ends of clauses with certain characteristic 
stress patterns. This is, of course, nothing more than a hint at a 
definition. A better one would have to be based on certain concepts 
which will be developed in chapter ten. 

The paradigmatic classes, nouns, verbs, adjectives, and pro- 

Outline of English Morphology 107 

nouns, also have their characteristic usages. For example, no noun, 
adjective, or pronoun can occur in the place a verb usually occu¬ 
pies in the predicate of a sentence. Conversely, a verb or a pronoun 
never occurs after the, though cither a noun or an adjective may. 
Such facts as these make it possible to recognize a noun, adjective, 
or verb in the base form, or indeed in any of the others when the 
full paradigm is not known. 

This is important because a very large proportion of English 
stems can be used in two or more parts of speech. For example, 
many, such as run, walk, nap, breakfast can be used as either nouns 
or verbs. That is, both the paradigm run runs ran run running 
and the paradigm run runs run’s runs’ occur. It is not possible to 
assign run in isolation to either class. But in contexts, such as I run 
a store, or There’s a run on the bank., there is no difficulty. Not all 
noun stems can also be used as verb stems, nor all verb stems as 
noun stems. If this were the case, the two parts of speech could 
not be distinguished. Nevertheless, the increasing use of words in 
other classes than those to which they originally belonged gives 
promise of drastic changes in English structure if present trends 

8.22 There are two basic processes of stem formation in 
English: (1) the addition of derivational affixes to roots or to stems 
of two or more morphemes, and (2) the combination of two or 
more stems to form compounds. A full description of these proc¬ 
esses in all their ramifications would be lengthy. We can only 
give a few illustrative examples of the problems involved: 

Stem formation of the first type is customarily described 
on the basis of the affixes used. Thus, for each affix is noted: 

( 1 ) the class or classes of stems (including roots) with which it 
is used, and any pertinent restrictions within the class or classes; 

(2) the class of stems produced; and (3) any morphophonemic 
changes in either the affix itself or the stem. For example, /-0/ is 
a suffix used with verb and adjective roots to form nouns: /gr6w0/ 
growth is from the verb root /grow/, and /d6pQ/ depth from the 
adjective root /diyp/. /—iy/ is a suffix which forms adjectives from 
noun stems: /gluwmiy/ gloomy from /gluwm/. It is used with some 
stems formed with /—G/; /filOiy/ filthy from /filG/ from /fawl/; 
/h 610 iy/ healthy from /hel 0 / from /hiyl/. 

None of the derivational affixes can be used freely with any 

108 Outline of English Morphology 

stem, even with any stem in the proper class. Some approach this 
rather closely, /-or/ ‘one who does’ forms nouns from a very large 
number of verb stems: /dtiwar/ doer, /rftytor/ writer, etc. It is a 
productive formation. That is, it is quite freely extended to new 
verb stems as the occasion arises. However, it is not used with 
all verb roots; *meaner ‘one who means’ or *seemer ‘one who 
seems’ do not occur and are rejected by many Americans as 
strange or impossible. At the other extreme, /—6/ is restricted to 
less than twenty stems. New formations arc not currently being 
made through the use of this suffix. In fact, the historic connection 
between some of the words containing /—6/ and the roots from 
which they were formed has been almost completely lost sight of. 
Not everyone would recognize the morphologic connection be¬ 
tween /d60/ death and /dhy/ die, and still fewer that between 
/fil0/ filth and /fawl/ foul. 

8.23 A first analysis will yield a large number of tentative 
morphemes with more or less similar meanings and functions. 
For example, /-liis/ -ness, /—itiy/ -ily, and /—0/ -th all have much 
the same meaning and are used in similar ways to form nouns 
from adjective stems. The question arises as to whether these can 
be considered as allomorphs of one morpheme. As between /-nis/ 
and /—Q/, the question is settled by a contrasting minimal pair, 
/w5HrmO/ warmth : /woHrmnis/ warmness. They are clearly to be 
kept separate. But what about /-nis/ and /-itiy/, or /-itiy/ and 


Actually, this question is of more theoretical than practical 
importance. A person learning English will have to learn what 
stems can take /-nis/ and which /-itiy/, whether they are allo¬ 
morphs of one morpheme or distinct morphemes. The problem 
here is very different from that with inflectional affixes. There is 
practical value in knowing that /-in/ in oxen is an allomorph of 
{-Zi}, for this tells us that oxen functions in the language in a 
similar way to boys, tables, etc. — that is, that we must say The 
oxen are .. . and not *The oxen is .. . 

8.24 If we take English words apart into their constituent 
morphemes, we find that the stresses present a problem. An isolated 
word typically has one and only one primary stress. Where in the 
word it occurs has no particular connection with the roots or 
affixes concerned, but seems to be a feature of the word as a whole. 

Outline of English Morphology 109 

To leave the stresses on the vowels on which they are found in the 
word is not a satisfactory division. 

An alternative method is to divide a word in the following 
manner: reading /riydfg/ - /' V + /riyd/ + /-iq/. /riyd/ is, of 
course, a root, and /-iq/ is an inflectional suffix. This leaves only 
/' ~/ to be accounted for. This combination of stresses is structur¬ 
ally significant for three reasons: (1) Reading is never said (at 
least in normal speech) with any other pattern of stresses, such as 
/*rlydlq/ or /*rlydiq/. (2) The same pattern occurs in numerous 
other words, among them /rfydSr/ reader, /gdwlij/ going. (3) Other 
patterns occur with the same constancy on other words, for 
example, /**'/ in /Sb$v/ above, /b!f5Hr/ before, etc. Indeed, 
minimal pairs can be found, /p6rm!t/ (noun) : /permit/ (verb) 
permit. The logical conclusion is that /' **/, /~ '/, and various 
others must be considered as morphemes. 

Do these morphemes have meanings? In pairs of nouns and 
verbs like permit, index, present, etc., they seem to. But in most 
instances this is not so clear. They are, however, comparable to 
la in I want ta go. In both cases (/to and /'” /) the structure de¬ 
mands these not obviously meaningful elements. Their function 
is to mark the internal structure of the utterance in which they 
occur. To indicates the relationship between want and go; /' “/ 
indicates that between /riyd/ and /-iq/. Both are morphemes for 
the reason that they form part of the morpheme system of the 
language; without them the system (as presently constituted) 
would be inoperable. 

8.26 The significance of stress morphemes is nowhere more 
evident than in compounding. At first sight this type of stem 
formation consists of nothing more than the juxtaposition of two 
stems; thus, green and house may be put together to form green¬ 
house ‘a glass house for growing plants.' But this is certainly not 
the whole truth, since the same items can be put together in the 
same order to produce green house ‘a house that is green.’ The 
difference is one of stress: /grlyn+h&ws/ has a morpheme /' + '/ 
which is characteristic of compounds; /griyn+haws ~ grlyn+hfiws/ 
has one of the morphemes /' +A / or / A + '/, which are characteristic 
of such constructions as adjectival plus noun or verb plus object. 
Many other instances of this contrast can be cited: bldck-bird: bldck 
bird, spit-fire : spit fire (as in It began ta spit fire.), grky hdund : grty 

110 Outline of English Morphology 

hdund. There are also many cases of compounds with this /' + '/ 
morpheme for which there are no such minimal pairs: bird-edge, 
bdth tab. 

An English word characteristically consists of a root and a stress 
morpheme, with or without derivational and inflectional affixes. 
Rather exceptionally there may be two or more roots, but there 
may be several derivational affixes. With one exception, {-ZiZ*}, 
no word has more than one inflectional suffix. Every word said in 
isolation normally has one and only one stress morpheme, though 
in context the stress morphemes ma}' be altered in certain regular 

These statements will not serve as a definition of a word, un¬ 
fortunately, because of certain complications which occasionally 
arise. The word is one of the most difficult concepts in English 
morphology to define, though in the vast majority of cases little 
question can arise as to whether a given sequence of morphemes 
is or is not a word. 




Some Types of Inflection 

9.1 A monolingual student expects that a second language will 
differ from his native speech in many details, but he is seldom 
prepared for the differences in basic structure which are commonly 
found. Most Americans who have studied some language in high 
school or college are only slightly more aware of the possibilities. 
French, Spanish, German, and Latin are all closely related to 
English and quite similar in many features. Moreover, they are 
commonly taught in such a way as to emphasize the similarities 
and suppress the differences. Were Chinese, Swahili, Aztec, or 
Navaho taught in our schools, they would serve much better to 
deepen the student’s linguistic perspective. Even a brief introduc¬ 
tion to these or any other non-Indo-European language will demon¬ 
strate that the languages of the world show remarkable variation 
in the most basic features. 

The diversity in languages is so great that it would hardly seem 
passible to say anything at all about the structure of languages 
in general. Yet, while very few features can be found which are 
common to all languages, a much larger number will be found to be 
very widespread. Throughout this chapter you must bear in mind 
that the features described are not necessarily universal. They are 
presented as indicative of the possible range of variation and as 
illustrative of frequent types of structure. 

9.2 The most generally useful method of describing the struc-. 


112 Some Types of Inflection 

ture of words is by analysis into morphemes and the description 
of the ways in which the morphemes can be combined. Always 
there are restrictions of various kinds on the combinations which 
can occur. These restrictions may affect the order in which the 
morphemes can be arranged; or there may be sets of morphemes 
which can never occur together in the same word; or certain classes 
of morphemes may be required to occur in certain circumstances. 
In addition there are frequently complex patterns of selection of 

In some instances the number of affixes used in a single paradigm 
is very large; or a single word may consist of a rather long series of 
morphemes. It is necessary to have some simple way of stating the 
complex combinations which can occur. This can often be done by 
classifying the morphemes into groups known as orders which are 
most conveniently designated by numbers. Thus, order 1 con¬ 
sists of all those suffixes which can occur only immediately after 
the root. Order 2 consists of those which can occur immediately 
after a morpheme of order I, or immediately after the root if no 
morpheme of order 1 is present, but never farther from the root 
than this. Order 3 consists of those which can occur only after 
roots or members of orders 1 or 2. Similarly, prefixes can be classed 
into orders. If both prefixes and suffixes occur, prefixes may be 
distinguished by the use of negative numerals. Thus, order -1 
would include those prefixes which occur only immediately before 
the root. Order -2 would consist of those which can occur only 
immediately before the root or a member of order -1. 

Only one affix of a given order can occur in a given word. If, 
for example, two members of order 1 should occur, then one would 
have to follow the other. By our definition, a member of order 1 
can only follow a root. Our classification of affixes into orders 
would be shown to be erroneous, and we would have to revise it 
to accord with the facts of the language. Orders are, therefore, 
mutually exclusive classes of morphemes occupying definable 
places in the sequence of morphemes forming a word. 

9.3 The utility of this method can be illustrated by the fol¬ 
lowing brief description of the Turkish verb. Turkish has very few 
prefixes, but a very extensive series of suffixes. The following list 
includes only the most important of those used with verbs. As it 
stands, it illustrates the salient features of Turkish verb structure; 

Some Types of Inflection 113 

to make it complete would add further complications and little of 

illustrative value. 

Order 1: /—il—/ 






Order 2: /-tir-/ 


Order 3: /-ma-/ 


Order 4: /-ir-/ 

‘habitual action’ 


‘ continuous action ’ 


‘future action’ 


‘obligatory action’ 

Order 5: /—<!*—/ 


Order 6: /-lar-/ 

‘third person plural actor’ 

Order 7: /-sa-/ 


Order 8: /-m-/ 

‘first person singular actor’ 


‘first person plural actor’ 


‘second person singular actor' 


‘second person plural actor’ 

Order 9: /-mi-/ 


Order 10: /-im/ 

‘first person singular actor’ 


‘first person plural actor’ 


‘second person singular actor’ 


‘second person plural actor’ 

9.4 There are various restrictions on the occurrence of some of 
these affixes beyond what is expressed by the classification into 
orders. These additional restrictions must be explained in notes: 
Only one of orders 6, 8, and 10 can occur. That is, these three orders 
are mutually exclusive. They cannot, however, be treated as a 
single order because of different arrangements when orders 7 and !) 
are also involved. Order 10 cannot occur with either order 5 or 
order 7. Conversely, order 8 can occur only with either order 5 or 
order 7 or both. 

Various other types of mutual relationships between orders of 
morphemes occur in other languages. Not infrequently, if a certain 
morpheme or any one of a certain class of morphemes occurs, it is 
obligatory for one member of some order to occur. This restriction 
is the converse of the mutually exclusive relationship which occurs 
between orders 6, 8, and 10 in the Turkish verb. No one of the 

114 Some Types of Inflection 

orders of affixes in the Turkish verb is required. A single word can 
have affixes from a considerable number of them, or no affix at all. 

9.5 In the absence of an}' morpheme of a given order, some 
specific type of meaning may be implied. Thus, in Turkish, in most 
circumstances, the absence of any member of either order 8 or 
order 10 implies ‘third person actor.’ This may be either singular 
or plural, unless /-lar-/ (order 6) is present, in which case it is 
necessarily plural. 

9.6 The following forms will illustrate the operation of the 
system just described. The affixes are arranged in columns to show 
their assignment to orders. At the top a list of the morphemes in 
each order is given for reference. 




3 4 5 6 



9 10 




ma ir di lar 



mi im 














a kirdi 



‘it broke’ 
b kirilmadilarmi 



ma di lar 


‘were they not 


c kirajaksan 





‘if you are going 

to break’ 

d dalisajakdim 


ajak di 


‘ I was going to 


c fialistirmalisin 





‘you ought to 



/ *kirili§di 


I * 1 



g *kirajaklarim 


ajak lar 


h *kirdisin 




i *kirsamali 




Some Types of Inflection 115 

The last four forms (with asterisks) are impossible. They are 
given only to illustrate the operation of the restrictions mentioned 
above. Because they are impassible, they arc meaningless, so no 
glosses are given. Example / has two morphemes of the same order. 
These are by definition mutually exclusive. If a word such as that 
given could occur, it would require assignment of /-il-/ and /-is-/ 
to different orders, but such words do not occur. (Note: /*kiriliSdi/ 
is not said to be impossible because /-il-/ and /—i§—/ are of the 
same order; rather, these two morphemes are described as be¬ 
longing to the same order because, among other things, such a 
form as /‘kirilisdi/ does not occur. Orders are a device to state 
restrictions rather than rules determining what forms arc to be 
prohibited or allowed.) Example g contains two morphemes which, 
though in different orders, are mutually exclusive. (Note: The 
difficulty does not occur because the meanings seem incompatible, 
but because of the nature of Turkish structure. Example d contains 
affixes glossed ‘future’ and ‘past.’ These might seem at first sight 
unreconcilable. But forms like /tfalisajakdim/ do occur and arc 
meaningful. In any case, it would be very dangerous to rest our 
analysis so thoroughly on glosses; they certainly do not give a full 
idea of the meanings of these affixes.) Example h is impassible be¬ 
cause with /-di-/ (order 5) the actor is expressed by a morpheme 
of order 8 (perhaps /—n—/) rather than one of order 10 (as /-sin/). 
In the example * the morphemes are all compatible, but a member 
of order 7 precedes a member of order 4. The same morphemes in 
the arrangement /kirmalisa/ would be acceptable and might be 
glossed ‘if he ought to break.’ 

9.7 Even with the abridgment which was made in this state¬ 
ment, something over three thousand verb forms are mathemati¬ 
cally possible from any given root. It is obviously inefficient to 
try to present a full paradigm; moreover, it is completely un¬ 
necessary, and it would be quite misleading if it were done. Some 
of the possible forms are quite common; others are used very 
rarely, if at all. But these forms, though rare, are possible, and 
would be understood when used. When the need for one of them 
arises, a Turk produces the proper form without realizing that he 
may never have heard that word before. Another Turk will under¬ 
stand it with equal readiness. The type of description just given 
is a formalized statement of the way a native speaker constructs 

116 Some Types of Inflection 

needed verb forms in the Turkish language. A paradigm giving a 
fixed repertoire of prefabricated verbs is unsuited to such a 

9.8 This statement of Turkish verb forms is seriously inade¬ 
quate in one respect. It mentions only one of the often numerous 
allomorphs of each affix. A full statement would have to list all 
the allomorphs and give rules for the correct selection under any 
possible conditions. Many of the morphemes have parallel sets of 
allomorphs with similar conditioning. It is, therefore, possible to 
make certain general morphophonemic statements which apply 
quite universally in the system. For example, the affixes of order 10 
have the following allomorphs conditioned by the vowel of the 

preceding syllable: 

1 sing. 

1 plur. 

2 sing. 

2 plur. 

after /i e/ 





after /ti 6/ 





after /i a/ 





after /u o/ 





Similar sets of allomorphs occur in all the other affixes which 
were listed in 9.3 as having the vowel /»/. (Some of them have other 
types of variations also.) This is an instance of vowel harmony. 
The whole system can be summed up in two quite general rules: 
Suffixes containing high vowels have — 

/i/ if the preceding syllable has /i e/ (front, unrounded) 


/u 6/ (front, rounded) 


/i a/ (back, unrounded) 


/u o/ (back, rounded) 

Suffixes containing low vowels have — 

/e/ if the preceding syllable has /i e ii 6/ (front) 

/a/ /iauo/ (back) 

Exceptions are not numerous. One is indicated in the listing. An 
affix in order 4 has the allomorphs /-iyor- ~ -uyor- ~ -iyor- ~ 
-uyor-/, with only the first vowel showing vowel harmony. 

9.9 The situation can be quite different from the relatively 
regular structure of Turkish. A description of a small part of the 
verb system of Cree (an American Indian language of Canada) will 

Some Types of Inflection 117 

illustrate a much more complex type of paradigm. In this language 
there are four classes of verbs. The following description is of 
part of the inflection of transitive animate verbs. These are verbs 
used with a subject and an object both of which are animate. 
(Animate is a gender class of nouns; every noun in Cree is either 
animate or inanimate.) A Cree verb exists in fifteen modes. The 
independent indicative and the conjunct indicative are two of the 
most frequently used. Complete paradigms of the endings for both 
these modes are given in the table on page 1 IS. 

Every transitive verb has both a subject and an object expressed 
in the ending. In the paradigm the meanings of these are indicated 
by code numbers. A hyphen follows the indication of subject and 
precedes that of object: e.g., 1-3 indicates ‘I.. . him/ whereas 
3-1 indicates ‘he . . . me.’ 

1 first person singular, ‘I’ 

2 second person singular, ‘thou’ 

3 third person proximate singular, the chief person spoken of, 
or the first one mentioned, ‘he,’ ‘she/ ‘it’ 

3' third person obviative, a person or persons spoken of, other 
than the chief character of the narrative, or other than the 
first one mentioned, ‘he,’ ‘she/ ‘it/ ‘they/ ‘the other’ 

3p third person proximate plural, ‘they’ 

2p second person plural, ‘you all’ 

12 first person plural inclusive, the speaker and the person 
addressed, ‘we/ ‘you and I' 

lp first person plural exclusive, the speaker and some person 
or persons other than the one addressed, ‘we/ ‘he and 1/ 
‘they and I’ 

Blank spaces in the table represent combinations of subject and 
object which do not occur. For example, ‘he . . . him’ would have 
to be either 3-3' or 3'-3. 3-3 is impossible because if there are two 
third persons they are distinguished, one as proximate and one as 

9.10 These paradigms are formidable. The whole set would be 
even more so. For the learner they represent a considerable load 
of sheer memorization. For the linguist they fail to show clearly any 
recurrent regularities of structure. Some simplification and sys¬ 
tematization would be desirable from both points of view. The 


118 Some Types of Inflection 




■a*?* r 

¥ - * f f '"y 
f T ? T ? T ? T f t 





. si A 

7'??77 i it 

fags f' n 

f <N Ki i , 

7 r 7 T 7 T ? T 7 t | , i l 


13 « 


•j •- 

TT i i 


£ — 


a — e. 

•r m 

4'i , 

. f n 1 

TT i 


i 1 i : i 


1 1 1 1 

T ? f 
, , ! A ! A i A 






- f *Pr 

.ti — 

I c 

.£ — .£ 04 


, , I 1 ! 1 ! 4 

TT i 

i i T i i 

T T i i 

1 1 1 1 

T '% 1 


I I 



l l TTT 



i TS 
i« S<n Si 

. . . P? 


| all 

. mi! 

•a i i i T T T T’T 


1 ! || 

g. I 1 i TTT T’T 

_ « 

I § I I I 

^ *- 









« V 

^ — 5 CO 
« CO « 

- a 5 ,5 c.,= 35 3 

K 5 f ?f 

: at 



rlV*?-? - 

c -=-i, i, -il 
.:* + & 1 

4 A ? i; 

C. Ol 

=!, § J. ^ 

Some Types of Inflection 119 

full set of Crce verb paradigms would not appear at first sight 
greatly different in complexity from a full set of Turkish paradigms. 
It is, however, possible to present the structure of Turkish verbs 
in a way that is both simpler to learn and more revealing of struc¬ 
ture. The same sort of analysis and reorganization seems to be 
needed in the Cree verbs. 

Only a little experimenting is necessary to demonstrate that the 
problem is vastly different. Turkish verb forms can be analyzed 
rather easily in terms of a series of affixes. The Cree endings cannot 
be neatly dissected into morphemes. For example, we would expect 
to find a morpheme marking the independent indicative, or one 
marking the conjunct indicative, or both. That is, there ought to be 
some morphemic difference between the two. It should be possible 
to identify this difference by recurrent contrasts between each of 
the 42 forms in one paradigm with the corresponding forms in the 
other. However, when the forms are compared it is found that the 
differences arc extremely various. The following is just a sample: 

Differences of one phoneme: 3-1 —ik : —it 

1-2 -itin : -itan 
3-2 -ik : -isk 

Differences of two phonemes: 1-3 -stw : ~ak 

3'-3 -ik : -ikot 

Greater differences: lp-3 -anan :-akiht etc. 

Differences of this sort cannot easily be summarized in terms of 
a single morpheme or a contrasting pair of morphemes. Attempting 
such a summary would necessitate describing a very long and 
complex set of allomorphs. Certainly no simplification can be 

9.11 There are some recurrent patterns that can be discerned. 
For example, comparison of all 3p forms with the corresponding 3 
forms suggests that a morpheme with the tentative allomorphs 
/-ak ~ -wak ~ -ik ~ -ok/ can be identified. It must be labeled 
“tentative” until the remainders of these forms can be analyzed 
satisfactorily. Unfortunately, such an analysis cannot be made, 
since most of the recurrent resemblances are limited to a part of 
one of the modes, or show such complications as to defy analysis. 
The paradigms as they stand arc unanalyzable. 

120 Some Types of Inflection 

Analysis can be carried a bit farther if some preliminary recon¬ 
struction is done. There are a number of morphophonemic changes 
which are known to occur in Cree: For example, a final /w/ after 
a consonant is dropped. Final /w/ can be added to certain forms. 
/*wa/ and /*wa/ sometimes change to /o/. Some of the forms 
with /o/ can be reconstructed by substituting /*wa/ or /*wa/. 
/«aa/ can be substituted for /a/ which might have been produced 
from /*aa/. lp-3p independent /-ananak/ can be reconstructed 
as /*-aan5nak/, and 3p-lp /-ikonanak/ as /*-ikwananak/; once 
this is done, they can be divided into /*-&-5nanak/ and /*-ikw- 
anfinak/. /-a-/ occurs in many of the forms in the upper right- 
hand half of the table of endings of the independent indicative; 
•it indicates that the nearer of the two persons involved is the sub¬ 
ject, and the farther is the object. For this purpose “nearer” means 
1 rather than 2 or 3, 2 rather than 3, and 3 rather than 3'. Similarly, 
/—ikw-/ occurs in many of the forms in the lower left half of the 
table. It indicates that the farther of the two persons involved is 
the subject, and the nearer is the object. After /-a-/ and /-ikw-/ 
are taken out, lp-3p and 3p-lp are alike, /-ananak/ must indicate 
that lp and 3p are both involved in some way, but without speci¬ 
fying in what way. It may be further divided into /-anan/ ‘lp is 
involved’ and /-ak/ ‘3p is involved.’ 

9.12 On the basis of such a procedure, almost all of the endings 
of the independent indicative mode can be analyzed. The result is 
the following list of morphemic elements: 

/-in ~ -w ~ -ayi/ 

‘1 or 2 is involved’ 

/-anan -an/ 

‘ lp is involved’ 


‘12 is involved’ 


‘2p is involved’ 

/-wa ~ -awa ~ -iw ~ 0/ 

'3' is involved’ 


‘3p is involved’ 

/-a ~ -it ~ -ima/ 

‘nearer is subject; farther is object’ 

/-ikw ~ 0/ 

‘farther is subject; nearer is object’ 

There is never any overt indication that 3 is involved, so no mor¬ 
pheme can be set up for this meaning. We may consider it as im¬ 
plicit when there is no indication of other participants. 

Some Types of Inflection 121 

The following morphophonemic changes have been assumed: 

/*ww/ becomes /w/ 

Final /*w/ after a consonant drops 
/*wa ~ *wa/ after a consonant becomes /o/ 

/*aa ~ *aa/ becomes /a/ 

/*ai/ ' becomes /e/ 

A complete statement would have to describe the distribution 
of the allomorphs. These are all morphologically conditioned, and 
the distribution is not easy to state simply. How this analysis 
works is best seen by comparing the reconstructed forms given in 
the table with the unanalyzed paradigm. The reconstruction is 
divided into morphemes. Below each is written a symbol to indi¬ 
cate its identity. 

9.13 There are three forms which are not analyzed. 3'-12 is 
identical with 3-1 p. We may assume that the distinction has been 
lost in this context, and that the 3'-lp form has been extended in 
meaning. Similarly, lp-2p and 2p-lp seem to have arisen as exten¬ 
sions of use of l-2p and 2p-l respectively. To treat them in any 
other way would immensely increase the complexity of the 

9.14 We have already pointed out one major failing of the 
method of analysis into morphemes. Several others are evident. 
For example, we listed /-ima/ as indicating ‘nearer is subject.' 
This is certainly the easiest way of treating it, if only the inde¬ 
pendent indicative is examined. But when the conjunct indicative 
is analyzed, /-im-/ must be treated as marking 3'. Such an analysis 
as has just been presented fails because it does not attempt to 
cover the totality of the data. This analysis is complex enough, but 
if it is extended to cover both modes, many additional and more 
intricate problems arise. Indeed, we may expect that the addition 
of still more of the fifteen modes will make the analysis still more 

Even more serious than the difficulty of analysis is the incon¬ 
clusive nature of the result. Even in the small part of the total 
system which was analyzed, there are numerous instances of very 
free conjecture. For example, /—ik/ ‘3-1’ was reconstructed as 
/*-ikw-w/. There was little warrant for either /w/. In short, 
any attempt to analyze these forms will inevitably take the 

122 Some Types of Inflection 

linguist out onto very thin ice. No one can be satisfied with the 

The analysis of Cree verbs into sequences of morphemes has a 
very different sort of usefulness than the comparable analysis of 
Turkish verbs. It is of very little direct use to a learner. A full 
morphemic analysis would be much more difficult to learn than the 
paradigms. Nevertheless, a partial analysis can be quite useful even 
to the beginner in providing some understanding of the paradigm. 
He will be assisted by learning the more obvious morphemes like 
/- a k ~ -ik ~ -ok/ ‘3p,’ or by understanding the contrast between 
‘nearer-farther’ and ‘farther-nearer’ forms. 

A detailed analysis does have both interest and value to a 
linguist. It gives a clearer insight into the structure than a list of 
forms ever could; it points out certain problems that may need 
further attention; it gives some basis for understanding some 
features of the history of the language. In short, it has more 
theoretical than practical value, though it does have some of 


9.16 The verbs of Loma (of Liberia) are quite different. The 
total paradigm includes only four forms, but there are two entirely 
arbitrary subclasses: 





Base form 









Recent past 





Far past 





The continuous is formed by the suffix /-sti ~ -zil/» and the recent 
past by /-g& ~ -a/- Since the stems which take /-sft/ also take 
/-g&/, and those that take /-zil/ also take /-a/, it is possible and 
useful to set up paradigmatic subclasses. 

9.16 A relatively short paradigm like this does not imply that 
Loma cannot express a number of distinctions of the sort that 
Turkish or Cree express through their larger assortment of verb 
forms. These four verb forms are used with various auxiliaries and 
with six sets of subject pronouns. Some of the distinctions which 
we might expect to be expressed in the verb form are expressed by 
the pronouns: 

Some Types of Inflection 


























w k 

































Loma is thus in very broad outline similar to English in that 
extensive inflection is replaced by a long series of syntactic con¬ 
structions. The various combinations of pronouns, auxiliaries, and 
verb forms express a considerable number of different verbal ideas. 
In English, comparable forms are constructed of various verbal 
auxiliaries and verbs. 

9.17 French has a moderately extensive paradigm of verb 
forms. The structure of many of the forms is such that analysis 
into morphemes, though possible, is not always simple and may 
not be very helpful. Four “regular” subclasses arc traditionally 
recognized. There are in addition a large number of “irregular” 
verbs. Some of these differ from the “regular” verbs in only minor 
respects. In others the divergencies are extreme. The “irregular” 
verbs often fall into small groups that are inflected alike. If, start¬ 
ing with the four “regular” subclasses, we arrange all the groups 
of like inflection in order of size, a characteristic distribution be¬ 
comes evident. This is shown in the figure. 

124 Some Types of Inflection 

This distribution is quite similar to that which was found in the 
English verbs described in Chapter 8. The largest of the French 
subclasses contains well over half of all the verbs of the language. 
More than one third of all the subclasses contain only one or two 
verbs apiece. From the point of view of the total vocabulary the 
“irregular” subclasses are quite unimportant. 

This is, however, by no means the whole picture. Many of the 
irregular verbs are quite common, and conversely many of the 
common verbs are irregular. Of the commonest verbs in the lan¬ 
guage, about one third are irregular, and these include a dispropor¬ 
tionate share of the most highly irregular. Among verbs of moder¬ 
ate frequency, only about one tenth are irregular. Among verbs 
which are rather infrequently used or distinctly rare, those with 
irregularities are a very inconspicuous proportion. The first sub¬ 
class is the productive subclass of French verbs. That is, most new 
verbs fall into this paradigm, and there is some tendency for other 
verbs to be conformed to it. This class includes about half of the 
commonest verbs, about four-fifths of the verbs of moderate fre¬ 
quency, and a still larger proportion of the verbs which are more 
rarely used. 

The practical effect of this is well known to any student of 
first-year French. The irregular verbs loom up as a major hurdle. 
There is one consolation; they will become less of a problem as he 
masters the commonest words and moves on. This same phenome¬ 
non is found in many other languages: the irregular forms are 
comparatively few in number, but often include many of the com¬ 
monest words in the language. This is so, partly because the 
commonest words most successfully resist the pressure toward con¬ 
formity with prevailing patterns. 

9.18 The several verb systems just sketched briefly indicate 
something of the range of variation of types of inflection. Three 
variables require comment: 

First is the number of forms in the paradigm. Turkish stands 
near one extreme with over three thousand. Cree is not far behind. 
English and Loma, with five and four respectively, stand near the 
other extreme. To complete the picture, it must be noticed that 
some languages have no inflection at all. This is the case in Malay. 
But in most languages there are at least one or two classes of words 
which are inflected. 

Some Types of Inflection 125 

Second is the complexity of formation. Turkish stands near one 
extreme in this matter; with only a few special restrictions, forms 
can be freely created by combination of affixes. Loma likewise has 
a very simple inflectional system, though here the short paradigm 
may be partly responsible. Near the other extreme is the verbal 
system of Cree. Here the combinations of affixes are so complex 
that it is hardly profitable, for practical purposes, to attempt to 
describe the formations. Fortunately, the division of words into 
stems and inflectional affix combinations is generally simple, so 
that a paradigm of one verb will serve readily as a pattern for the 
inflection of another. English, though it has a short paradigm, 
shows a relatively high degree of complexity. 

Third is the number of inflectional subclasses. Turkish has only 
one. That is, the inflection of all Turkish verbs can be compre¬ 
hended under one scheme of affixes and morphophonemic rules. 
Loma has only two; within each the inflection is quite simple, but 
it is necessary to know to which subclass any given stem belongs. 
English and French have more numerous subclasses; these are 
merely arbitrary divisions, classes of verbs having similar mor¬ 
phology. Cree subclasses are quite different. The verbs described 
were transitive animate verbs — that is, those which can and must 
have an object and whose object must be an animate noun. There 
are also transitive inanimate verbs, which must have an inanimate 
object; animate intransitive verbs, which have no object and whose 
subject is animate; and inanimate intransitive verbs, which have 
no object and whose subject is inanimate. The differences are not a 
matter of “meaning,” but of usage. Frequently English transla¬ 
tions will flatly contradict these distinctions and cannot, of course, 
be used to distinguish the verb classes. The subclasses depend on 
the syntactic patterns of the language, which arc largely arbitrary, 
and on the two subclasses of nouns, which are also largely arbitrary. 

9.19 With so much variation in the extent and nature of in¬ 
flection in various languages, the form of analysis and description 
used must be suited to the language at hand. There is sometimes a 
tendency to force the description of a language into the pattern 
most familiar from past experience. For generations, Latin gram¬ 
mar supplied the pattern. Frequently, the familiar Latin parts of 
speech were recognized, whether they were there or not. Often the 
forms were fitted into the paradigmatic framework of Latin. Until 

126 Some Types of Inflection 

quite recently, English and American school children were required 
to learn such paradigms as the following, which is given with the 
Latin model: 

Singular nominative the boy pucr 

genitive of the boy pueri 

dative to the boy puero 

accusative the boy puerum 

ablative from the boy puero 

vocative 0 boy! puer 

and similarly through the plural. 

The result of this sort of grammar is twofold. It gives the stu¬ 
dents the impression that grammar is essentially a specialized type 
of formalized nonsense, of no practical value, though traditionally 
part of the educational process. A direct consequence is the popu¬ 
larity of foreign language courses that promise “no grammar.” 
This is, of course, ridiculous. A language is a systematic structure; 
to learn a language is to learn this structure. Any description of this 
structure is grammar. 

A second result is to blind the observer to many features of the 
language which are properly the concern of grammar, but which 
are not usually treated in the traditional Latin grammar. In mak¬ 
ing the grammar of English as nearly like that of Latin as possible, 
resort must be had to conflation as seen in the paradigm above, to 
some measure of distortion, and also to the complete neglect of 
features which cannot be made to conform. Because of the domi¬ 
nance of basically Latin concepts, the grammar of English has been 
until very recently much less well known than that of many much 
less used languages. The latter owe their superior descriptions to 
the fact that the first approach had been made by linguists with 
fewer prejudgments about grammar. 

The bondage of English grammar to Latin patterns is being 
broken, and with the break has come a period of rapid advance in 
our understanding. Part of the credit for this belongs to modern 
descriptive linguistics, which has brought to English a deeper 
perspective gained from the examination of a wide variety of 
language structures. But at least as much of the credit accrues to 
students of the English language as such, who by closer attention 

Some Types of Inflection 127 

to the material they are studying have been led to much the same 
discoveries more or less independently. 

The opposite, but ultimately identical, error has also been made. 
Some linguists have become so enamoured of a fixed pattern of 
analysis originating in revolt against Latin patterns that they 
have used it where it is no more appropriate than the other. 
Paradigms have been over-used by traditionalists and justly re¬ 
volted against. But in some languages the listing of forms in 
paradigms is the most feasible and useful technique of description. 

9.20 The wide variation in the number of forms in paradigms 
has led some into false conclusions about the functional adequacy 
of various languages. That Turkish has several thousand verb 
forms, whereas Malay has only one, docs not imply that Turkish 
can express more different shades of meaning. There are other 
techniques by which a language can express fine differences of 
meaning. It would be necessary to examine the whole system of 
both languages before any such judgment could be made. But 
any reasonably close examination demonstrates immediately that 
it is senseless to make such a comparison: these two languages, like 
any two languages, may be expected to structure content in such 
different ways that direct comparison is quite meaningless. 

It is, however, a safe generalization to say that all languages are 
approximately equally adequate for the needs of the culture of 
which they are a part. This is most particularly true when the 
cultures are relatively static. When they are not, all languages seem 
to be able to adjust with approximately the same lag, as culture 
changes impose new requirements on the communication system. 

The evaluative comparison of languages can be a gross form of 
ethnocentrism and is usually utterly sterile. That some African 
language might be an inadequate medium to describe a World 
Series game is to be expected. Incidentally, Shakespeare’s English 
would do little better. Nor, of course, is English satisfactory as a 
vehicle for the description of some intricate facet of African culture. 
Even with the highly developed special terminology of the an¬ 
thropologist, difficulty is experienced; English must be eked out 
with numerous technical terms from the language of the com¬ 
munity under scrutiny. But this proves little, since the most ob¬ 
vious deficiencies are in vocabulary, and new words can be created 
rapidly in any language as the need arises. 





10.1 Grammar is conveniently divided into two portions: mor¬ 
phology and syntax. Syntax may be roughly defined as the 
principles of arrangement of the constructions formed by the proc¬ 
ess of derivation and inflection (words) into larger constructions 
of various kinds. The distinction between morphology and syntax 
is not always sharp. In some languages such a definition of syntax 
is reasonably useful. In others it poses serious difficulties. But a 
more satisfactory discrimination cannot be found to cover lan¬ 
guages generally. Nevertheless, in spite of the vagueness of the 
limits of syntax, the principles which will be discussed below apply 
very widely in languages and are often of great usefulness even 
where syntax is least sharply set off from the remainder of gram¬ 

10.2 Both the nature of the problem and the general approach 
to its solution are best presented through an example. For this 
purpose we will start with a sample of written English. Our first 
approach will be by rule of thumb, without the check of rigorously 
established method. This approach will serve primarily to indicate 
the nature of the problem. In this preliminary discussion, we will 
arbitrarily disregard those features of the language which are not 
indicated in the spelling (stress, pitch, and transitions), but we will 
make use of the indicated word divisions, thus sidestepping the 
issue as to what is a word. In this particular example, nothing is 


Syntax 129 

falsified by this practice but in some other examples the spelled 
word divisions may be misleading. 

10.3 The old man who lives there has gone to his son’s house. 

This utterance contains twelve words. We may, as a first 
hypothesis, consider that each of them has some statable relation¬ 
ship to each other word. If we can describe these interrelationships 
completely, we will have described the syntax of the utterance in 
its entirety. 

As we attempt to do so, we will soon discover that the nature 
of the relationship varies widely from pair to pair. For example 
old and man have a clear, direct relationship which is relatively 
easily stated. Old and house present no such clear direct relation¬ 
ship, and any discernible connection is quite complex and seem¬ 
ingly less interesting. One might conclude that this is merely 
because old and man are near to each other in the utterance, 
whereas house and old are widely separated. This may have some 
truth in it, but it is certainly not the whole story, for we find no 
particularly close relationship between there and has , though they 
are contiguous, while a much closer relationship is sensed to exist 
between man and has. Since the pairs of words vary so widely in 
the closeness of relationship which a native speaker feels to exist 
between them, the description of the interrelation of each with 
each other would seem to be a very inefficient procedure. More¬ 
over, it could be very cumbersome. This utterance of twelve words 
would require the analysis of 66 relationships; an utterance of 
one hundred, 4,950. 

10.4 As a second possibility we might start by marking those 
pairs of words which arc felt to have the closest relationship. We 
will also lay down the rule that each word can be marked as a 
member of only one such pair. Something like the following might 
be the result: 

The old man who lives there has gone to his son’s house. 

At a second step in our procedure, let us assume that these pairs 
of words function in the utterance as single units. There is some 
reason to suggest that they do, since we can replace any of these 
by a single word and get a sentence which, though different in 

130 Syntax 

meaning, seems to be in some sense similar in structure. For 


old man 


lives there 

has gone 


his son's 










10.6 If this procedure is valid, there is no reason why it cannot 
be repeated as many times as may be useful. Something like the 
following might result: 

The old man who lives there has gone to his son’s house. 

V_ > \ _/ V_ > V_ > 





These steps may be paralleled by the following series of utterances: 


old man 


lives there 

has gone l 


his son’s 






















By this procedure we have reduced our example successively 
from twelve to eight, then six, then four, and finally only two items. 

10.6 Much the same result can be attained by proceeding in the 
opposite direction. A native speaker might be asked to mark the 
most fundamental cleavage in the utterance. This would probably 
be between The old man who lives there and has gone to his son’s 
house. (That this is so is involved in our earlier observation that 
there and has, though adjacent, seemed to have no very clear, direct 
association.) This process can be repeated with each portion, until 
the ultimate divisions consist of single words. (This is merely the 
limit of our present concern. The method could be extended by 
dividing at a still deeper level between, say live and -s. This would 
carry us out of syntax and into morphology.) The result would 
probably be something like this: 

The | old : man |i who ] lives : there | has ; gone | to || his ; son's | house. 

Syntax 131 

This result is identical with that obtained before. However, 
with some other utterances, the two methods might lead to similar 
rather than identical results. 

10.7 The procedure which we have just sketched will be useful 
to us, if it serves as a framework within which all the relationships 
of the utterance can be effectively and economically described. 
Let us consider that between old and house. We feel that the rela¬ 
tionship here is not direct. Our analysis shows that they are, in fact, 
about as remote from each other in the structure as can be. Any 
relationship which exists between them arises from the fact that 
each participates in the formation of an item which ultimately is 
related in a definable way to that containing the other. We may 
symbolize it as follows: 

old ■* -►man 

old man-* - ywho lives there 

The^———+-ol^nan who lives there 

The old man who lives thcrc+—+has gone to his son’s house. 

has gone -—*-lohis son’s house. 

to^——his son’s house. 

.. ,/\ 

his son’s-*—*-house. 

The heavier line is intended to indicate the most direct relation¬ 
ship between old and house. This is certainly complex and rather 
tenuous. It confirms our earlier impression that the interrelation¬ 
ships of the two words were hardly worth describing except for our 
desire to state the syntax of the utterance completely. This dia¬ 
gram, however, also established this important fact: the interrela¬ 
tionship of old and house, unimportant in itself, is nevertheless 
describable in terms of a chain of relationships each of which 
individually seems significant. We can, therefore, attain our objec¬ 
tive of complete description by a proper selection of the relation¬ 
ships to be described. 

10.8 When properly done, the method of describing the struc¬ 
ture of utterances in terms of successively larger constructions and 

132 Syntax 

their relationships proves to be quite generally feasible and useful. 
In many instances it provides a simple framework within which 
everything significant can be effectively and efficiently described. 
In some instances where this is not so, it is found that this pro¬ 
cedure generally disposes of most of the significant features 
efficiently, leaving only a small residue which must be treated by 
other methods. 

As the demonstration above was carried out, the process of 
uniting items into larger items was somewhat haphazard. The 
whole procedure was based on the uncontrolled intuition of a 
native speaker. If several Americans had been asked to do the 
same thing, each working independently, the results would not 
necessarily have been identical. One person working with several 
utterances might analyze similar utterances in noncomparable 
ways. Such an erratic procedure might simplify description in some 
measure, but it could never be as effective as a procedure which 
would necessarily attain comparable results with comparable mate¬ 
rials. Moreover, the method as sketched would be useless to a 
linguist confronted with a language for which he lacked a native 
speaker's “feel.” 

It should be possible to establish a method of finding the best 
possible organization of anj r given utterance and of insuring com¬ 
parable results with comparable material. This is the basic problem 
of syntax. In the ensuing paragraphs some of the factors involved 
will be discussed briefly. Unfortunately, the methodology has not 
as yet been completely worked out in a generally applicable form. 
Moreover, some of the best approximations to a general theory are 
beyond the scope of an introductory text. 

10.9 Further discussion will require certain definitions: 

A construction is any significant group of words (or morphemes). 
Thus in the example we have been discussing, the whole utterance 
is a construction. So is the old man who lives there or old man. 
But there has is not, since the two words have no direct connection. 
Neither is man, since it contains only one word. On a syntactic level 
lives is not a construction; but on another level it is a construction 
consisting of two morphemes, live and -s. 

A constituent is any word or construction (or morpheme) which 
enters into some larger construction. Thus, in the example we have 
been discussing, each of the words is a constituent. So likewise are 

Syntax 133 

old man and the old man who lives there. However, there has or man 
who is not a constituent. Neither is the utterance as a whole, since 
there is no larger construction of which it is a part. 

Note that all but the smallest constituents are constructions and 
all but the largest constructions are constituents. The two terms 
apply equally to a very large number of items. Which term is used 
of a given item depends on our interest. If we are concerned with 
the item as a part of a larger whole, it is a constituent; if as a whole 
composed of smaller parts, it is a construction. 

An immediate constituent (commonly abbreviated IC) is one of 
the two, or a few, constituents of which any given construction is 
directly formed. For example, the old man who lives there and has 
gone to his son’s house are immediate constituents of the utterance. 
Old man is an IC of old man who lives there, but not of the utterance 
as a whole. The ICs of a given construction are its constituents on 
the next lower level. Those on any still lower level are constituents 
but not immediate constituents. 

Of these concepts the most important is the immediate con¬ 
stituent. The process of analyzing syntax is largely one of finding 
successive layers of ICs and of immediate constructions, the de¬ 
scription of the relationships which exist between ICs, and the 
description of those relationships which are not efficiently described 
in terms of ICs. The last is generally of subsidiary importance; 
most of the relationships of any great significance are between ICs. 

10.10 The basic method for determining the ICs of any construc¬ 
tion is that of comparing samples. For example, let us consider 
what are the ICs of his son’s house. We will continue to operate 
with the written forms, assuming that the words are established 
constituents. Four ways of dividing our example are possible: 

his | son’s house, his son’s | house, /us[ son’s 1 house (with a discon¬ 
tinuous constituent his . . . house), and his | son's | house (with 
three ICs). The problem is to select which of these is preferable 
and to establish a rule which will enable us to arrive at an equiva¬ 
lent result with all comparable samples. 

If the construction being examined had only two constituents, 
only one method of division would be possible, and there would be 
no problem. This is the case, for example, with old man. We may 

134 Syntax 

find some two-word construction which is judged to be directly 
comparable with his son’s house. This would be some construction 
which can occur in similar environments and which is alike in all 
those features which we shall see to be used to mark syntactic 
relationships. Such a one might be John’s house. In such a case, 
the most obvious analysis is that his son's of the first example is 
equivalent to John’s of the other. On this basis we will divide as 

his son’s house 
John’s house 

To base our conclusion on a single such comparison might be 
dangerous but for the fact that many others can be found, and the 
weight of evidence will clearly favor such a division as against 
any other possibility. 

10.11 Another possible procedure can be based on a series of 
comparisons such as the following: 

The old man who lives there has gone to his son’s house. 

The old man who lives there has gone home. 

If this comparison is correct, we may consider that it establishes 
that to his son’s house is a constituent, since home must be. We 
cannot draw the conclusion that it establishes to his son's house 
as an IC, since we have no reason to believe that home is an IC of 
the sentence in which it occurs. Any sequence of words which can 
be shown to be equivalent to a single word is a constituent, since 
all single words may be assumed to be constituents. If, as is the 
case here, the single word is also a single morpheme, the conclusion 
is quite safe. If we proceed in this way we should be able to identify 
all of the constituents in any utterance, and once this is done the 
identification of the ICs and immediate constructions is easy. 

10.12 That the structure of a given utterance can be worked 
out in this manner is not sufficient to explain the fact that the 
native speaker confronted with a sample of his language imme¬ 
diately senses some of this same structure. He would seem to 
depend on some sort of markers of the ICs of an utterance. That 
such is the case can be shown by the native speaker’s reaction to 
such an utterance as: 

The iggle squigs trazed wombly in the harlish goop. 

Syntax 135 

Only three words of the nine are recognized, and there arc no other 
utterances containing the unfamiliar ones with which to make 
comparisons. Nevertheless, to any native speaker of English, the 
structure is quite clear, however obscure the meaning may be. It 
is almost unmistakably marked by the three familiar words the and 
in the and the four word fractions -s, -ed, -ly, and -ish. 

These items do not directly mark structure as such, but they 
indicate certain facts about specific constituents. For example, -ish 
is commonly (not universally!) a derivational suffix forming ad¬ 
jectives or words with similar syntactic functions. This scrap of 
evidence is confirmed by the fact that harlish occurs in a position 
where such a word is to be expected. This in turn, together with 
the position of in the, suggests that goop most probably is a noun, 
and that in the harlish goop is a constituent of a certain quite com¬ 
mon type. This suggests to the native speaker that the utterance 
is in some way comparable to such as the following: 

... in the harlish goop 

He lived in the red house. 

I read it in the big booh. etc. 

So, in a sense, these markers serve to identify other utterances with 
which the one under consideration can be compared. However, 
they do so through the intermediary of certain other facts about 
the constituents. We must therefore note in detail what these 
devices are. 

10.13 The first of these is so subtle that it is easily overlooked. 
This is word order. The significance of this is plain if the words 
are scrambled: 

Goop harlish iggle in squigs the the trazed xoombly. 

The native speaker sees no structure. This is not simply because 
ICs arc no longer adjacent. Such is not a necessary condition. 
Consider the utterance: 

What are you looking for? 

What . . . for and are . . . looking are constituents. They are not, 
however, merely discontinuous; they are discontinuous in a set and 
regular way. To make them continuous will just as surely destroy 
the intelligibility as to jumble the words in any other way: 

What for are looking you? 

136 Syntax 

Word order is probably the most fundamental of syntactic 
signals in every language. It is also one of the most complex. 
Every other device that will be discussed implies some use of word 
order. Other syntactic signals can seldom be described adequately 
without close attention to word order, nor can word order be 
described except in the context of the total syntactic structure. 

10.14 A second universal syntactic device is that of constituent 
classes. This is an extension of the notion of parts of speech which 
was discussed briefly in Chapter 8. A constituent class is any group 
of constituents (words or constructions) which have similar or 
identical syntactic function. They are frequently based on the 
paradigmatic word classes of the language. For example, English 
has four of the latter: noun, pronoun, adjective, and verb. Each of 
these has some syntactic characteristics in addition to the in¬ 
flectional characteristics by which they were defined. Each can 
therefore serve as the base for a constituent class. Adjectives quite 
characteristically occur in such an environment as: 

The . .. man . .. 

The good man . . . 

The loll man . . . etc. 

We may therefore define a class of constituents which will occupy 
this place. It will include the adjectives, and hence the items in this 
class can be called adjectivals. (We will adhere to the convention 
of labeling all syntactic classes with terms ending in - al .) The class 
of adjectivals will include all adjectives, certain uninflccted ad- 
jective-like words (such as beautiful), and certain constructions 
mostly having adjectivals for one constituent: 

The generous 

man . . . 

The most awkward 

man . . . 

The intolerably ugly 

man . . . 

The most exceptionally brilliant 

man . . . etc. 

Similarly we can define a class of nominals including nouns and 
various equivalent constructions; pronominals including pronouns 
and certain equivalent constituents, e.g., the 'party of the first part 
of legal jargon; and verbals including verbs and various construc¬ 
tions occurring in the same contexts as verbs. The syntactic func¬ 
tions of pronominals are so nearlj r the same as those of nominals in 

Syntax 137 

English that it is probably better to consider pronominals as a 
subclass of nominals. 

Similarly we must recognize certain other constituent classes 
not based on inflectional classes. Among these arc adverbials and 
prepositionals. These include the traditional parts of speech known 
as adverbs and prepositions, but are wider in that they also include 
constructions. For example, in regard to is a prepositional, though 
not a preposition in the traditional definition. 

Other languages have comparable systems of syntactic classes, 
but we must expect that they will differ widely in the details of 
the system. 

10.15 In describing the syntax of English it is also desirable to 
recognize certain construction classes. These are types of construc¬ 
tions which cannot be assigned to any single constituent class. A 
typical example is the prepositional phrase. This always consists of 
a prepositional and a nominal. Since each of these constituents can 
be either a single word or a rather complex construction, preposi¬ 
tional phrases can be superficially quite different, varying from 
in truth to in regard to our wholesaler's last large shipment. This 
variation is only superficial; all show the same basic structure. But 
as constituents of larger constructions, prepositional phrases can 
function in many different ways, so that they do not form a con¬ 
stituent class. They may be adverbials as in He came in a big hurry. 
or a special type of adjectival as in The man in the car . . . 

Another type of construction class is the subject-predicate sen¬ 
tence. This is a construction which has as its immediate con¬ 
stituents a subject and a predicate. A subject is a nominal, a 
pronominal, certain types of phrases, or a clause. A predicate is 
a verbal or various larger constructions involving verbals. Neither 
of these descriptions is a definition. The defining characteristic is 
that subjects and predicates serve as immediate constituents in 
the formation of sentences. These definitions are both circular 
and inexact. A more adequate set would have to be built up 
through a long chain of definitions of the various types of con¬ 
stituents and constructions occurring in English. This would 
require much more space than can be given here. The important 
notion, however, is that sentences have a definable regularity of 
formation, and that this is expressible in terms of ICs. It must 
be noted that subject-predicate sentences are not the only type in 

138 Syntax 

English, but merely the mast frequent. The other types, for ex¬ 
ample that including The more, the merrier., can be analyzed in 
comparable ways on the basis of ICs and certain formal relation¬ 
ships between them. 

10.16 The examples discussed up to this point were deliberately 
presented in orthography rather than in phonemic transcription. 
This demonstrates that enough signals are commonly present in 
spelled English to make sentences intelligible. But this is not 
always true. In an early draft of this text I wrote: What thinking 
Americans do about language is . . . When I reread the script this 
sentence seemed irrelevant until I realized that the written form 
is ambiguous; I had read it as meaning ‘What is done about lan¬ 
guage by thinking Americans is . . whereas I had written it to 
mean ‘What thinking about language is done by Americans 
is . . .’ If this sentence is read aloud, there is no ambiguity. It 
can be read to carry either meaning, but not both simultaneously. 

The ambiguity rests in different ways of organizing the words 
into constituents. The intention was that A/nericans do about 
language should be a constituent, not thinking Americans. There 
is nothing in the written form to signal this. But in speech these 
relationships are clearly marked by stress and intonation. Perhaps 
the most usual way of reading this sentence in either sense will 
be with two intonation contours, the first of which ends with /—*/. 
The position of this /—»/ will always mark a major division in the 
utterance. If the /->/ occurs at the end of thinking, only the in¬ 
tended meaning is possible. If there is no /—>/ here but there is one 
after Americans, the other meaning is indicated. 

To mark constructions is the chief function of both the stress 
and intonation systems of American English. Stress patterns most 
commonly operate at the word level, as was noted in chapter eight, 
or tie together very intimate groups of words. Intonation is used 
almast entirely on the syntactic level and commonly marks rela¬ 
tively large constructions. Both systems consist of morphemes 
which stretch over sequences of other morphemes (roots and 
ordinary affixes) and serve to mark their organization into con¬ 
structions. In spoken English the stress and pitch contours are 
additional ICs of the constructions in which they occur. 

/ 2 5iy + 6wld + m^n + huw +3 livz-t'6r 2 ~' 2 haz + g5Hntuw + iz +3 s6nz' f haws 1 ^/ 

Syntax 139 

There are three ICs in this sentence, a subject The old man who 
lives there, a predicate has gone to his son’s house, and an intonation 
pattern consisting of two contours /232—»231\/. The stresses 
mark structures within the subject or the predicate respectively, 
and so are not ICs of the sentence. The stress and pitch contours 
are not represented in written English, but their place is taken by 
two other devices: the spaces marking the conventional word divi¬ 
sions, and the punctuation system marking certain conventional 
syntactic units. 

10.17 Syntax has long been one of the least satisfactorily 
handled aspects of the structure of languages. This appears to be 
largely because of failure to give sufficient attention to stress and 
intonation or to the equivalent features of the language concerned. 
(Both pitch and stress can function in other languages in ways very 
different from the ways they function in English, but there is 
probably always in any language a system of some kind which 
marks constructions as pitch and stress do in English.) As the 
major features of English intonation have become known in recent 
years, great progress has been made in the analysis of syntax. 
Rigorous description of sentence structure no longer seems an 
unattainable goal, as it did only a decade or so ago. Some pre¬ 
liminary work along the same lines in other languages has in¬ 
dicated that similar results are distinctly possible in these, too. 

10.18 There are two other syntactic devices which deserve 
mention. Each is of very minor significance in English, though 
much more important in certain other languages. The first of these 
is government. This means that certain inflectional forms arc used 
primarily to signal the place of the word in the syntactic structure. 
When nouns are involved, the most familiar instance, these special 
inflectional categories are called cases. Case is also common in 
pronouns; for example, the difference between I and me or he and 
him is one of case. Such contrasts arc found in English only in the 
pronouns. Each of these case forms is restricted to certain struc¬ 
tural positions. For example, me and him occur with prepositionals, 
with most verbals in the predicate, etc. / and he in standard 
English do not occur in these places. The distribution of these two 
cases follows somewhat different rules in some other dialects, and 
in the speech or writing of many Americans is not very consistent, 
much to the consternation of purists. 

140 Syntax 

Because these forms are restricted to certain syntactic positions, 
they serve to signal the structure of the utterance. Thus in I saw 
him. the forms I and him rather than me and he assist in marking 
the relationships. However, in English, word order alone is used 
for this purpose with most norainals. Paul saw Mary, is fully as 
clear, though there is one less structural marker. The result of this 
is that Americans rely very little on the case forms even when they 
are available. If a group of Americans are instructed to correct 
such sentences as *Mc saw Paul, and *Mary saw he., a majority 
will make them I saw Paul, and Mary saw him. rather than Paul 
saw me. and He saw Mary. This may be taken to indicate that, 
when the facts of word order and case form are in conflict, native 
speakers of English will consider word order as the more significant. 

10.19 The situation is quite otherwise in some other languages. 
In Latin, most nouns show case forms which mark the sentence 
structure much more clearly than do the cases of English pronouns. 
Moreover, they have greater functional significance in Latin than 
in English. For example, Paul saw Mary, might be translated: 

Paulus Mariam vidit. Mariam Paulus vidit. 

Paulus vidit Mariam. Mariam vidit Paulus. 

Vidit Paulus Mariam. Vidit Mariam Paulus. 

Any one of the six would be clear, since the inflectional affix -us 
marks Paulus as subject, and -am marks Mariam as forming, with 
vidit, the predicate. Though all are intelligible, not all arc equally 
“good.” There are strong preferences for one as against the others, 
but at various periods in the history of the language these prefer¬ 
ences have been different. 

It is sometimes said that because of the highly developed in¬ 
flectional system of Latin, word order was unimportant. This is a 
gross overstatement. In every language word order has important 
syntactic functions. The example just quoted is exceptional in 
Latin in allowing nearly absolute freedom of word order. Every 
language has some definable instances of rigidly fixed word order 
and some definable freedom of word order. All that can fairly be 
said is that in Latin word order is less important as a syntactic 
device than in English. It is still of great importance. 

10.20 Another device to indicate structure is concord. This 
means that certain words are required to take forms which cor- 

Syntax 141 

respond in certain ways with certain other words. There is very 
little concord in present-day English. The clearest instance is with 
the words this and that , which are required to show concord in 
number with any noun with which they are associated. Thus we 
say that boy but those boys, and similarly this boy but these boys. 
These are clear instances of concord, but they have relatively little 
functional value in English because they arc isolated vestiges. 

Latin has a much better-developed system of adjective-noun 
concord. Every adjective must agree with its noun in three cate¬ 
gories: number, gender, and case. 

filius bonus 
fdii boni 
puella bona 
puellarum bonarum 

‘the good son’ (sing. masc. nominative) 

‘of the good son’ (sing. masc. genitive) 
‘the good girl’ (sing. fem. nominative) 

‘of the good girls’ (plur. fem. genitive) etc. 

Moreover, Latin concord serves a syntactic function, since occa¬ 
sionally only the concord indicates the immediate constituents. 
Consider the following examples: 

fil ips, domini bonm 

‘the good son of the master’ 

filius domi ni boni ^ 

‘the son of the good master’ 

Concord may involve various inflectional categories. In Hebrew, 
nouns in apposition must show concord in gender, number, and 
definiteness. The latter refers to the presence or absence of a 
prefix usually translated ‘the,’ or to certain equivalent conditions. 
For example: 

/m$lek gaaddol/ 

/hamm&ek haggaaddol/ 
/malk&a godoolaa/ 

/molaakiim godooliim/ 
/hammalaaklim habboruuklim/ 

‘a great king’ (sing. masc. indef.) 
‘the great king’ (sing. masc. def.) 
‘a great queen’ (sing. fem. indef.) 
‘great kings’ (plur. masc. indef.) 
‘the fortunate kings’ (plur. masc. 

10.21 Concord between the head noun of the subject and the 
verb or other head word of the predicate is also common. By head 
is meant that constituent which seems to serve as the center of any 
construction. It will typically belong to the same syntactic class 
as the construction of which it is the head. Here also, there is some 

142 Syntax 

trace in English. The {-Z*} form of the verb is a concord form 
indicating that the subject is third person singular. It occurs only 
in the present. 

Latin shows a comparable type of subject-verb concord much 
more fully developed. All verb forms have distinct singular and 
plural forms. 

Filius vidit. ‘The son saw.’ (singular) 

Filii vidcrunl. ‘The sons saw.’ (plural) 

This type of concord differs from the adjective-noun concord of 
Latin in that gender and case are not involved. 

Hebrew has a similar type of concord. In this instance both 
gender and number are involved: 

/zaak&r hamm&ek/ ‘The king remembered.’ (masc. sing.) 

/zaakoraa haminalk&a/ ‘The queen remembered.’ (fem. sing.) 
/zaakoruu hammolaakiim/ ‘The kings remembered.’ (masc. plur.) 

In Hebrew a predicate can be either a verb or a noun or various 
longer constructions. When the predicate is a noun, the same type 
of concord occurs as when it is a verb. That is, there must be 
agreement in number and gender. Since the subject is always 
definite and the predicate indefinite, a subject-predicate sentence 
is clearly distinguishable from a construction of two nouns in 
apposition, by the different type of concord. 

/gaad6ol hamm61ek/ ‘The king is great.’ (a sentence) 

/m61ek haggaadbol/ ‘The great one is king.’ (a sentence) 

/hamm&ek haggaad6ol/ ‘The great king’ (not a sentence) 

10.22 Concord is one instance of a relationship which may exist 
between constituents other than immediate constituents of an 
utterance. For example, in the Latin sentence Filius bonus esl. 
‘The son is good.’ gender concord occurs between fdius and bonus. 
The immediate constituents of the sentence, however, are fdius 
and bonus est. The relationship may be diagrammed: 










Some Inflectional 

11.1 Expression and content are equally fundamental aspects of 
language. The last six chapters have sketched some of the phe¬ 
nomena of grammar, that part of the expression system directly 
related to content. Throughout the discussion, occasional refer¬ 
ence was made to the content categories expressed. In this chapter 
we will discuss some of these more explicitly and attempt to point 
out some of the more important characteristics of such units in 
language structure. The topics discussed have been selected be¬ 
cause they are commonly associated with the inflectional systems 
of languages, and the illustrations will be drawn mostly from use 
in this connection. This must not be taken to imply that any one 
of the categories mentioned will necessarily have anything to do 
with the inflectional system of any particular language. Languages 
differ as widely in the categories expressed in inflection as they do 
in the morphologic structure by which they are expressed. 

11.2 In some languages inflectional categories very familiar to 
speakers of European languages are entirely lacking. For example, 
in many languages nouns or noun-like words are characterized by 
inflection for number. This is by no means either universal or 
necessary, though our English background leads us both to assume 


144 Some Inflectional Categories 

it to be normal and to be puzzled that a language can exist without 
it. But many do, the most familiar example being Chinese. Words 
semantically equivalent to numerals, or other terms for quantity, 
do exist in Chinese and can be used when called for. English like¬ 
wise has similar words which can be used whenever the number is 
particularly significant. The difference between the two languages 
is that the expression of number in Chinese is optional; in English 
it is compulsory. Every English noun is required by the language 
structure to be either singular or plural. In many cases this is 
informationally unimportant. In a large part of the instances where 
number is significant, numerals or other expressions of quantity 
are added, making the singular-plural distinction in the noun 
itself redundant. 

Many have said that the lack of number inflection in Chinese 
was a mark of inadequacy of the language. With perhaps more 
justification one could say that the distinction in English is of 
little value, being in most instances either irrelevant or redundant. 
Neither conclusion is defensible. Number has a significant function 
in English grammar, whereas Chinese has a grammatical system 
in which number has no such part. Both languages are workable 
and each has its own systematic structure composed of just such 
details, most of which cannot be individually justified. When a 
brick house is to be built, it does not matter whether this brick or 
that brick is used, but some brick must be used. Number is one of 
the bricks in English structure; Chinese uses others. 

11.3 Number is familiarly thought of as a contrast between 
one category indicating a single individual and another indicating 
two or more. These are traditionally designated singular and 
plural. These names are intended to suggest the “meanings” of 
these categories. 

Probably the category of number has a more obvious and direct 
connection with demonstrable contrasts in the world of experience 
than any other inflectional category of English. Because of this 
seeming objectivity, number is an ideal example to demonstrate 
that all such categories are, at least in part, arbitrary. Some others 
are very highly arbitrary with very little demonstrable connection 
with observable phenomena. English number (in the singular- 
plural sense) is a structuring imposed on experience by English 
patterns; it is part of language, not of nature. 

Some Inflectional Categories 145 

There is an old story of a man who was asked, presumably by a 
grammarian, whether pants was singular or plural. His reply was, 
“Well, mine are plural at the bottom, and singular at the top.” 
Ultimately, the confusion, which many others have also felt, rests 
not so much in the shape of the garment as in the grammar of 
English. The object named is as clearly one entity as, say, a shirt 
or a coat. This does not matter; by a convention of English, pants 
is plural. Interestingly enough, this is not an isolated case; com¬ 
pare trousers, breeches, shorts, slacks, etc. This whole group of 
words are grammatically plural with no evident semantic justi¬ 

11.4 English nouns fall into two major classes with regard to 
the semantic value of number. They may be referred to as count 
nouns and mass nouns. No discussion of the meaning of singular 
and plural can be realistic without carefully distinguisliing between 
these. The two classes are also syntactic subclasses of nouns, since 
they differ markedly in the use of the articles. In general, singular 
mass nouns use the articles in the same way as do plural count 
nouns. For example, singular count nouns can be freely preceded 
by a, whereas mass nouns can be only in very special contexts. 
Singular mass nouns can be preceded by /s3m/ some ‘a quantity 
of’; this word is only used before plurals of count nouns. (/s5m/ 
some, though spelled the same, is a very different word mean¬ 
ing ‘a certain’ or something of the kind. It can be used much 
more freely.) 

The concords of verbs and of this and that depend solely on 
number, without regard to the contrast of mass and count nouns. 
The distinction between these subclasses of nouns is purely ar¬ 
bitrary. Consider rice and beans. Both refer to articles of food con¬ 
sisting of numerous small particles. Yet one is a mass noun and 
the other a count noun. The contrasts and resemblances in usage 
may be seen in the following examples: 

Mass noun: 

Rice is good for you. 

This rice is good. 

I choked on a grain of rice. 
*A rice . . . 

/s5m rfxys/ 

Count noun: 

Beans are good for you. 
These beans are good. 

I choked on a bean. 

*A grain of beans . . . 
/s5m bfynz/ 

146 Some Inflectional Categories 

Not only is the distinction between words like rice and beans 
arbitrary, but various dialects of the language differ at some /sdm/ 
points. Molasses in standard American is a singular mass noun like 
rice . In some other dialects it is a plural count noun like beans. 
Compare the following: 

Molasses is good for you. Molasscs are good for you. 
(standard) (some colloquials) 

I have never heard a singular *a molass, but this does not affect the 
case. I suspect that if a situation should arise which would elicit it, 
speakers of certain dialects would readily use it. 

11.5 Moreover, in all dialects certain words arc used either as 
count nouns or as mass nouns, always with a difference of meaning. 
In these cases, the difference is frequently signaled by the use of a 
before the count noun. Compare: 

A piece of iron A piece of an iron 

The mass noun refers to a material, the count noun to a tool. The 
difference is perhaps most clearly seen in such an utterance as: 

Thai piece of an iron is not a piece of iron; it's the wooden handle. 

The semantic relationship between the two may be rather obscure. 
Soldering irons arc customarily made of copper; we have had 
coppers made of iron-zinc alloy; and nickels contain almost no 
nickel. Even when such apparent contradictions do not need to 
be reckoned with, there is no way of predicting one meaning if 
the other is given. 

11.6 The singular of a mass noun, in the simplest case, refers to 
some quantity, usually undefined and frequently not countable, of 
some substance. The plural of a mass noun usually refers to a 
number of kinds or species of the substance. Thus metals does 
not imply several discrete occurrences of objects, but of a number 
of kinds of substances. In other instances, this formulation does 
not fit so clearly, but the meaning seems to be more nearly this 
than the traditional statement of the meaning of singular and 
plural. Consider the beauties of poetry. This certainly does not refer 
to any collection of discrete occurrences of things. It can be de¬ 
bated to what extent it has any sense of a number of kinds of 
beauty (that is, to what extent beauties parallels metals in mean- 

Some Inflectional Categories 147 

ing). These details need not detain us. Their full discussion can be 
either very involved or very profitless or both. What is important 
to note is that the category of plural in English gathers together a 
rather diverse assortment of concepts. All these have one thing in 
common; they contrast with another assortment of concepts which 
we call “singular.” That is, the unity within the category is purely 
a feature of the linguistic system of the language which arbitrarily 
sets these two in contrast and imposes the requirement that every 
noun be assigned to one or the other. The singular-plural contrast 
is common in languages. We must, however, expect that there will 
be considerable differences of detail or even of rather broad outlines 
between the assortments of concepts which various languages 
bring together into each of these categories. 

11.7 The possibilities of number distinctions are not exhausted 
with singular and plural. Some languages distinguish two or more 
plurals on the basis of distinctions which arc not inflectionally 
significant in English. For example, Kru, a language of Liberia, 
has a singular and two plurals. One of these refers to any chance 
assortment of two or more of the objects referred to. The other 
refers to a group of objects which are in some way related. Thus 
the English term men might be translated in' two different ways. 
One would refer to a chance group, the other, perhaps, to a number 
of men from the same tribe. One form to be translated ‘books’ 
might mean any odd assortment, the other a number of volumes 
from a set. 

Other languages make distinctions of the same sort that English 
does, but in greater detail. Many languages have three numbers: 
singular, dual, and plural. Dual refers to two of a kind. In such a 
system, plural applies to three or more. Less frequently there may 
be singular, dual, trial, and plural. 

Finally, there may be a system that combines some aspects of 
several types of distinctions. At one stage in Hebrew there were 
singular, dual, and plural. Dual referred to two objects which were 
members of a pair. Plural referred to three or more, or to two 
objects which were not members of a pair. Thus /yaadayim/ 
‘hands’ would refer to the two hands of a single person; /yaadiim/ 
‘hands’ would refer to any three or more hands, or to, say my 
hand and your hand, but not to my two hands. 

11.8 Gender is another category which is quite common in 

148 Some Inflectional Categories 

nouns. In English it is not richly developed. The gender of an 
English noun is defined solely in terms of the pronoun substitute, 
he, she, or it, which may be used in its place. Typically, gender 
involves not only substitution but also concord. Indeed, probably 
the best definition of gender is as a set of syntactic subclasses of 
nouns primarily controlling concord. 

Languages wliich have gender as a grammatical category vary 
widely as to number of genders. French, Hebrew, and Hindi 
have only two. Latin, Russian, and German have three. Some 
languages have more than a dozen. 

In European languages there is some correlation of gender with 
sex. This is reflected in our traditional labels, masculine, feminine, 
and neuter. However, the correlation may be exceedingly loose. 
The names of many sexless objects are assigned to either masculine 
or feminine, even in those languages which have a neuter. Of 
course, in a two-gender language like French, every noun must be 
assigned to either masculine or feminine. Rather more frequently 
than is commonly realized, male or female beings arc referred to 
by neuter nouns, and occasionally male beings are named by 
feminine nouns and female beings by masculine nouns. Gender is 
in large part a linguistic classification of nouns into arbitrary 
groups for syntactic purposes. Nevertheless it is not wholly 
arbitrary. This is shown by the high degree of agreement that will 
be found if, for example, monolingual speakers of German arc 
asked to assign genders to loan words. The content structure of 
the language seems to gather a number of disparate semantic 
categories together into three genders. It does so in a way that is 
in part arbitrary and in part systematic, at least to the extent that 
the native speaker can sense a proper place for a new word which 
will coincide with that selected by a fellow speaker. 

In many languages gender categories have nothing whatever to 
do with sex. A common type is that which distinguishes between 
animate and inanimate. This is the case, for example, in the 
Algonquian languages such as Cree. (Compare the verb inflection 
discussed in 9.9fT.) The animate class basically includes all persons, 
animals, spirits, and large trees. But the following are also arbi¬ 
trarily animate: tobacco, corn, apple, raspberry (but not straw¬ 
berry), feather, kettle, snowshoe, smoking pipe, etc. 

Many African languages have highly developed gender systems 

Some Inflectional Categories 149 

controlling concord. An example is Bariba, a language of French 
West Africa and Nigeria. The following noun classes are found 
(each example is given with an adjective to show the concord): 

1. /dum baka/ 

2. /kp&> bakaru/ 

3. /boo bako/ 

4. /d5n5n bako/ 

5. /yam bakam/ 

6. /tarn bakasu/ 

7. /gad bakanu/ 

‘a big horse' 
‘a big stone’ 
4 a big goat’ 

4 a big fire’ 

4 a big space’ 
4 a big yam’ 

‘a big thing’ 

Classes 3 and 4 have the same forms of the adjective. They can, 
however, be clearly distinguished by the concord forms of the 
'that’: /g6/ with class 3 and /wi/ with class 4. 

These seven genders arc only very loosely connected with any 
discernible differences in meaning. For example, all nouns referring 
to persons are in class 4, but so are also such words as ‘fire,’ 
‘ulcer,’ ‘mouth.’ 

11.9 Genders are primarily syntactic categories, but they may 
also have inflectional significance. In Bariba there are a number of 
different allomorphs of the plural affix. In some instances the 
plural can be predicted if the concord class (gender) is known. 
For example, all nouns of class 2 form the plural by adding /-nu/. 
Conversely, if the plural is known to be formed by /-nu/, it may 
be predicted that the noun belongs to gender 2 or 3. Class 3, 
however, includes nouns that form their plurals by adding /-nu/ 
or /-su/. There are a few nouns, mostly in class 1, which are 

A similar situation is familiar to students of Latin and Greek. 
For example, in Latin the first declension (a paradigmatic sub¬ 
class) is largely composed of feminine nouns. There are, however, 
a few masculines, including agricola, ‘farmer,’ poeta ‘poet,’ and 
nauta ‘sailor.’ The second declension is largely composed of 
masculine and neuter nouns, but contains a number of feminines, 
including fagus ‘beech,’ pinus ‘pine,’ and taxus ‘yew.’ That is, 
in spite of numerous exceptions, there is a positive and significant 
correlation between gender and inflection. 

11.10 In nouns, gender is commonly an inherent feature of each 
stem. That is, nouns are not inflected for gender, but each noun 

150 Some Inflectional Categories 

has a characteristic gender. In those languages with a well-de¬ 
veloped concord system, adjectives are usually inflected for gender; 
that is, no adjective has an inherent gender, but may be inflected 
to produce a form for each gender. In many instances this is a 
useful basis for distinguishing between these two parts of speech. 
It can be conveniently used in Latin and Swahili and many other 

Content categories are not inherently either inflectional, deriva¬ 
tional, or associated with the roots, even in a given language, but 
may be associated in different ways with the different parts of the 
expression system of the language. 

11.11 Person is a common category in verbs and pronouns. 
English verbal inflection for person is very rudimentary, so we 
will start our discussion from a consideration of the pronoun root 
forms. Eight sets of forms were listed in 8.18 as personal pronouns. 
Of these, who is a special case and will not be discussed here. The 
remaining seven are ordinarily thought of as belonging to one of 
three persons, each of which is said to occur in both singular and 
plural. Actually, some questions can be raised about this inter¬ 

The third person certainly has a singular and plural in English. 
These forms have much in common sjm tactically and semantically 
with singular and plural nouns, and can often be directly substi¬ 
tuted for them. The usage of pronouns parallels all the vagaries of 
the noun plurals. Where are my pants? They were right here. : Where 
is my shirt? It teas right here. In some situations they can substitute 
directly for a construction composed primarily of third person 
singular pronouns. He and she came. They soon went. All these 
considerations support treating they as the plural of he she it in 
much the same sense as dogs is the plural of dog. 

The second person presents a different situation. In most dialects 
there is absolutely no distinction whatever between the so-called 
singular and the so-called plural. You is used indiscriminately and 
always with the unmarked form of the verb. This is in contrast 
with the third person, where the differentiation between singular 
and plural pronouns is paralleled by concord forms of the verb 
(in the present). With no structural basis whatever, it is question¬ 
able procedure to use two labels, “second person singular” and 
“second person plural.” Such labels are justified in those dialects 

Some Inflectional Categories 151 

which do actually have two forms, frequently something like 
/ytiw/ : /yuwoHl/. 

In the first person the situation is even more complex. Two 
forms occur in all dialects. However, the distinction between them 
is seldom comparable to that between singular and plural in nouns, 
or to the distinction in the third person pronouns. / means, 
roughly, ‘the speaker.’ If it had a plural it would be expected to 
mean ‘the speakers.’ This is certainly not an ordinary meaning of 
we. It might occur in a choric reading, but this would be unusual 
and artificial. I do not remember ever having heard we used as a 
plural of I in any instance that was spontaneous and unrehearsed. 
While it is probably convenient to continue familiar terms, it is 
important to realize that “first person plural ” can be nothing more 
than a convenient designation for we; it is not a description. 

11.12 The commonest meaning of we is ‘the speaker and some¬ 
body else.’ In English the “somebody else” is literally anybody. 
In some other languages a distinction comparable to that which 
English makes between second and third persons is made in the 
“somebody else.” In the Cree paradigms of 9.1) one category refers 
to the speaker and the hearer; this is usually called the inclusive 
first person. Another refers to the speaker and someone other than 
the hearer; this is usually called the exclusive first person. Such 
a distinction occurs in many widely scattered languages, though, 
of course, there are differences of detail from language to language. 

In Cree, both forms of the first person plural are secondary, 
however. They arc formed by combinations of the following cate¬ 
gories. (The forms cited are used in noun inflection to indicate the 

/ke-/ ‘the hearer is involved’ 

/ne-/ ‘the speaker but not the hearer is involved’ 

/o-/ ‘neither speaker nor hearer is involved’ 

Alone these will be translatable by ‘your’ (singular), ‘my,’ and 
‘his.’ The latter involves no distinction of sex-gender, but is ani- 
ihate rather than inanimate. 

/-en5n/ ‘the speaker and someone else are involved’ 

/-waw/ ‘two or more but not the speaker are involved’ 

152 Some Inflectional Categories 

In combination the following meanings are expressed: 

/ke- -enfin/ 'our (inclusive)’ 

/ne- -enan/ ‘our (exclusive)’ 

/ke- -waw/ ‘your (plural)’ 

/o- -waw/ ‘their’ 

11.13 The same Cree paradigm includes forms illustrative of 
another type of person distinction, that between proximate and 
obviative third persons. Both refer to someone other than the 
speaker or hearer, but they distinguish between two such persons. 
Roughly, the proximate refers to the nearer one, the main char¬ 
acter in the narrative, or the one mentioned first. The obviative 
refers to the lesser character or to the one mentioned second. This 
is an inflectional category in both nouns and verbs, and the refer¬ 
ences are kept more or less clear by concord. For example: 

/kitotSw/ ‘he (prox.) talks to him (obv.)’ 

/kitotik/ ‘he (obv.) talks to him (prox.)’ 

/okimaw/ ‘chief (prox.)' 

/Okinawa/ ‘chief (obv.)’ 

/iskCw/ • ‘woman (prox.)’ 

/iskewa/ ‘woman (obv.)’ 

/okimaw iskewa kitotew/ ‘The chief talks to the woman.’ 
/okimawa iskew kitotik/ ditto 

/okimaw iskewa kitotik/ ‘The woman talks to the chief.’ 
/okimawa iskew kitotew/ ditto 

The sentences translated alike are of course not identical in 
meaning, since they may imply some difference in the place of the 
two individuals in the total narrative. Nevertheless, the primary 
function of the distinction is to identify the relationships between 
nouns and verbs. This is shown clearly in these sentences, which, 
though admittedly artificial in some respects, do illustrate the 
Cree method of marking certain syntactic relationships. There are 
numerous complexities in the uses of these distinctions, which, 
though presumably clear to a Cree, can be very perplexing to a 

11.14 English makes limited use of contrasts comparable to 
that in Cree between proximate and obviative. One common case 
is the use of the former : the loiter. Frequently the use of this : that 

Some Inflectional Categories 153 

parallels the Cree system in some respects. Both these devices are 
used with the same rather erratic shifting from one form to another 
which the student thinks he observes when he starts reading Cree. 
An American seldom has any difficulty following the story through 
a tangle of this and that because the usages follow established 
English conventions. 

The significant difference between these devices and the contrast 
in Cree between proximate and obviative is that, while English can 
distinguish in various ways between two characters in a narrative, 
it is not necessary to do so. In Cree, however, the proximate : ob¬ 
viative dichotomy is compulsory. This sort of nonconformity 
between the categories of languages imposes one of the major 
difficulties in translation. Consider the following sentences. 

James and John had a fight. He got a black eye. 

To translate this into Cree, one would have to decide who got the 
black eye. One of the two names ‘James’ and ‘John’ would have 
to be marked as obviative, since there cannot be two proximates in 
one context. Then ‘he’ would have to be marked as either proxi¬ 
mate or obviative in accordance with our decision. In Cree it is 
difficult to be ambiguous or to dodge the issue. In English am¬ 
biguity at this point is possible, though one can be as specific as 
he likes. If, however, the sentence were James and Mary . . ., 
English could not easily be ambiguous, but would identify the 
reference of the pronoun by using cither he or she. Cree would have 
to distinguish here also, but by the same device as before. But in 
certain other contexts, Cree could be ambiguous where English 
could not. Languages differ not as much in what they can express 
as in what they regularly do and must express. Translation must 
frequently add to the meaning by making distinctions which are 
required by the structure of one language but not specified in the 
text being translated. 

11.15 Most languages have words or affixes which are used to 
specify the particular instance intended; these are sometimes called 
demonstratives. In English there are two, this and that. The basic 
distinction between the two is a matter of proximity. This points 
out the thing which is nearer. That refers to the one which is 
farther away. The same contrast is found in the adverbials here 
and there and in the obsolescent words hither : thither and hence : 

154 Some Inflectional Categories 

thence. Of course there are numerous complications in the way this 
distinction is actually applied, but whenever the two are in con¬ 
trast, some form of this difference can be found. 

Other languages have similar contrasts but distinguish more 
degrees. This is the case in some English dialects which have a 
three-way contrast between here, there, and yonder, and between 
this, that .. . there, and that. . . yonder. Instances have been re¬ 
ported of several more gradations of this sort. Latin has three 
demonstratives, but the distinction involves another contrast. 
Hie generally means ‘near the speaker,’ iste can mean ‘near the 
hearer,’ and ilc ‘remote from both.’ Translation difficulties can 
easily arise. Hie is usually translated ‘this’ and *lie ‘that,’ but iste 
must be translated ‘this’ or ‘that’ depending on the context. In 
addition, of course, differences in the details of application of these 
contrasts in the two languages can produce other complexities. 

Another type of distinction which is commonly made is that 
between objects which are visible and those which are not visible. 
Since visible objects arc frequently nearer than invisible ones, the 
distinction is usually translatable by ‘this’ and ‘that.’ But such a 
translation may badly obscure the sense that was intended or even 
be contradictory to it. 

These examples will serve to indicate that demonstratives can¬ 
not be satisfactorily defined by translation glosses. This is, of 
course, true of many other linguistic contrasts, but is particularly 
troublesome with items of this sort. Only copious illustration of 
usage can ever convey the meanings of such words or morphemes. 
To be most effective these illustrations must be chosen with great 
care, and preferably should include examples of contrasts. 

11.16 Somewhat similar to the demonstratives are the articles. 
These are, however, much less widespread in the languages of the 
world, many tongues lacking them completely. The English articles 
arc customarily considered to have a pointing-out function similar 
to, but less forceful or precise than, the demonstratives. Actually, 
this function of the articles is much less prominent or important 
than is supposed. The use of a or the is almost entirely controlled 
by the syntax of the utterance rather than by the meaning. The 
articles are important structural signals with numerous and rather 
complex functions. Since they are primarily structural devices, it 
is not surprising that languages show extreme variation in the 

Some Inflectional Categories 155 

matter of articles. Many languages have none, others only one, 
others several. Apart from this the usages may be different not only 
in detail but also in broad outlines. 

The treatment of the articles of one language as equivalents of 
those in another has sometimes led to serious misunderstanding. 
For example, it has been said that Hebrew thought is less abstract 
than English, because, where an American might say Gold is good., 
a Hebrew speaker would say /toob hazzaah;\ab./, allegedly to be 
translated as ‘The gold is good.’, the latter being a less general 
statement than the regular English form. The fallacy is patently in 
equating the prefixed article /haC-/ with the, so that /hazzaahuab/ 
must be ‘the gold’ in contrast with /zaahaab/ ‘gold,’ and then 
reading English distinctions into the resulting translation. The 
reason there is no the in Gold is good, is that gold is a mass noun, 
and is signaled as such by the absence of the. With an English mass 
noun the particularizes. In Hebrew there is no contrast to cor¬ 
respond with the English mass noun : count noun. Therefore the 
argument about Hebrew being less abstract is irrelevant. Moreover, 
in a sentence of the pattern of /t6ob hazzaahaab./, /haC-/ is used 
as a structural signal to mark the subject. It has nothing whatever 
to do with abstractness or concreteness in a context such as this, 
though in some other contexts it may have functions somewhat 
comparable to those of English the. The whole alleged contrast 
fails, because /haC-/ is required by Hebrew structure in such a 
general statement, while the omission of the is required by English 
structure in an equivalent statement. 

Another instance in which the English article serves as a struc¬ 
tural signal is seen in the contrast a bowl or vessel : a bowl or a vessel. 
The first implies that bowl and vessel are synonyms and no contrast 
between the two is intended. In the second, the intention is to 
contrast the two and imply that if the object is a bowl, it is not also 
a vessel. Such a contrast is not inherent in the a as such, but in the 
different structural relationships which the presence or absence of 
the a signals. Such a contrast may be marked by radically different 
means in various other languages. 

11.17 Another category which is of frequent occurrence in 
languages is determined by the social status of some person or 
persons concerned in the conversation. This may express itself in 
various ways. In Tibetan, for example, there are many pairs of 

156 Some Inflectional Categories 

words with similar or identical meaning except that one form is 
used only when speaking to or of a person of high social status. 
The following samples will give some idea of the range of meanings 
involved and the extent of differences in expression: 














European languages commonly have some such distinction, 
though it is usually restricted to pronominals. For example, in 
English, your honor is a special pronominal used in addressing a 
judge. In German there is a much more widespread difference in 
the second person pronouns. Du is used in addressing children, 
servants, relatives, intimate friends, God (in public prayer), etc. 
Sie is used in formal situations, in addressing strangers, chance 
acquaintances, or superiors, etc. This distinction extends to the 
inflectional system, since the verbs show different concords, du 
bist: Sie sind ‘you are.’ 

The patterns of such usages can be extremely complex, and may 
be conditioned by the status of the person spoken to, of the speaker, 
or of the person spoken of; or the relative status of two or more of 
these. There may be several grades distinguished, so that the 
number of forms involved may be considerable. 

Americans not infrequently resent such patterns in another 
language, feeling that such conventions reflect a social attitude 
that the}' do not care to share or perpetuate. This may be so, or 
the distinctions may be largely formal with little connection with 
social discriminations. In either case, the American student of such 
a language will do well to learn to use the distinctions in the way 
they are used by native speakers whose status is comparable to his 
own. Failure to do so will usually be interpreted as evidence of bad 
manners, bad grammar, or ignorance, rather than as a testimony 
to equalitarianism. Moreover, language reform, if it is indeed 
needed, must originate from within the community, not from 
strangers with an imperfect knowledge of the speech habits of the 
language. A foreigner should be very cautious in proposing or fol¬ 
lowing any departure from prevailing patterns. 

Some Inflectional Categories 1 57 

11.18 Throughout the discussion in this chapter certain cate¬ 
gories have been treated in such a way as to imply that they are 
associated primarily with certain parts of speech. While it is prob¬ 
ably true that number, for example, is most often basically a 
category of nouns, this is not necessarily the case. For example, in 
Quileute (Oregon) both nouns and verbs have plurals. /a 9 t’Sit/ 
‘chief’ (singular) : /a’a’t’Sit/ ‘chiefs’ (plural), /<51a xali/ ‘I leave 
him' (singular): /6’ela-xali/ ‘I leave him often’ (plural), /6-xwal/ 
‘he carries water’ (singular) : /o-Vxwal/ ‘he carries water often’ 
(plural). The two verbal examples are exactly comparable in their 
formation to the noun cited. All three are formed by the infix 
/-’V-/- The latter two are not plural verbs in the sense that they 
are verb forms used with plural actors. Instead, they indicate the 
plurality of the action expressed. 

There is no grammatical category that is necessarily associated 
with any particular type of word. Each language has its own pat¬ 
terns. As between closely related languages (like most European 
languages) the patterns may be quite similar, and the similarities 
may be reinforced by conventional grammatical statements. As a 
result we are accustomed to think of certain categories as “always” 
expressed in certain ways. But languages of other relationships, 
associated with other cultures, may be fundamentally different, 
not only in how they express things, but in what they express and 
how. they associate in a sentence structure the various concepts 
which are mentioned. No single pattern can be taken as normative. 
One must carefully avoid the tendency to think of the conventions 
of his own language as logically necessary or as inherently more 
reasonable or convenient than those of others. 

chapter j 


! 12 

The Phoneme 

12.1 In Chapter 2 the consonants of English were examined and 
found to be twenty-four in number. Among them was one which 
we chose to represent by the symbol /k/. This is heard, among 
other places, in the English words key, sky, and caw. This result 
was attained by two processes, one explicitly discussed and the 
other more or less implicitly assumed. The first was that of finding 
instances in which this sound in one of its occurrences was de¬ 
monstrably in contrast with various other sounds. Thus the pair 
key and lea was put forth to establish /k/ and /t/ as contrasting 
phonemes. Some such pair can be found (not all were mentioned in 
the text) to establish /k/ as distinct from each of the other twenty- 
three consonants of English. 

The implicit process was to assume that the sound which we 
have labeled /k/ in key /kly/ is somehow “the same” as those 
which we labeled /k/ in ski /sidy/ or caw /k5n/. This seems ob¬ 
vious enough to an American. But it is not necessarily obvious to 
a person of a different linguistic background. An Arabic speaker, 
for example, might object that the sounds in key and caw are quite 
different, though he would probably accept the identification of 
those in key and ski. Conversely, a speaker of Hindi would protest 
at the identification of that of key with that of ski, but would 
probably accept the identification of those in key and caw. This 
step in the analysis seems to depend heavily on the linguistic back- 


The Phoneme 159 

ground of the observer. Obviously, if a linguist is to make an 
accurate phonemic analysis of any language, the procedure must 
somehow be freed from dependence on the native language patterns 
of the analyst. 

The original method of Chapter 2 depended on a definition of 
the phoneme which was adequate for the purpose at that point. 
However, it provided only a criterion for determining when two 
sounds are different, and overlooked entirely the problem of de¬ 
termining when two sounds are alike. For this we relied on the 
feeling of a native speaker for his own language. We must therefore 
start by redefining the phoneme from a more comprehensive view¬ 
point. In this we will follow up the idea that was merely hinted 

at in 2 . 21 . 

12.2 A phoneme is a class of sounds. For example, the /k/ in 
key is easily demonstrated to be different from those in ski or caw, 
as the latter are from each other. But this is by no means the full 
extent of variation within one phoneme. Similarly, several other 
phonemes vary so widely that no significant phonetic description 
can be made without mentioning the variation. There is no English 
phoneme which is the same in all environments, though in many 
phonemes the variation can easily be overlooked, particularly by 
a native speaker. But since these patterns arc not the same in any 
two languages, the foreigner is frequently struck by the differences 
that the native speaker does not hear. 

12.3 Perhaps the clearest way to comprehend this aspect of the 
phoneme is through a brief (and somewhat superficial) description 
of the process by which a child learns to hear and reproduce the 
phonemes of his own language. The human vocal apparatus can 
produce a very great variety of different sounds differing from 
each other in numerous features. At first, the child attaches no 
significance to any of these and babbles randomly, using a very 
wide selection of these possible sounds. After a while he learns to 
distinguish between certain parts of his repertoire and discovers 
some utility in certain sequences of sounds. None of the latter are 
pronounced with any high degree of precision or consistency, but 
nevertheless some distinctions become established and at this 
point speech has begun. The process is not one of learning to pro¬ 
duce sounds, but of distinguishing between sounds. 

My daughter rather early learned to distinguish between labial 

160 The Phoneme 

and non-labial stops. Later she learned to distinguish between 
voiced and voiceless stops. However, the contrast between /t/ and 
/k/ was established much later. There was a long period when 
enough other phonemic contrasts were in use to make her speech 
intelligible, at least to her parents, but when /t/ and /k/ were not 
distinguished. [t]-like sounds were somewhat commoner than [k]- 
like, but the latter occurred, as did various intermediate varieties. 
Thus cake was usually something which impressed adults as /t6yt/, 
but occasionally as /kdyt/ or /t6yk/ or even /k6yk/. These several 
pronunciations sounded different to adults, but apparently were 
all alike to her. That is, from her own point of view there was one 
voiceless non-labial stop /T/ which might be pronounced as [t] 
or [k] or various other sounds, and cake was pronounced /T6yT/. 
Of course, take, Kate and Tate (all of which were in her vocabulary) 
were pronounced alike, that is, with the same range of variation, 
and hence all were confused. After some time she discovered the 
distinction and with increasing precision sorted out the four pro¬ 
nunciations and assigned each to its proper usage. When this had 
become as regular and consistent as it is in adult speech (we all 
make occasional slips!), her old phoneme /T/ had given way to 
/t/ and /k/. She had progressed one more step in acquiring the 
adult phonemic pattern of English. 

Note that the process was merely that of dividing the total range 
of voiceless non-labial stops into two. Each continued as a range 
of variation, that is, as a class of sounds. Larger classes were re¬ 
placed by smaller, but single sounds never entered into the pic¬ 
ture. For this reason, if for no other, phonemes must be classes of 
sounds, but as we shall see there are other cogent reasons also. 

12.4 The incentive for learning to distinguish between /t/ and 
/k/ was primarily the need to distinguish between such words as 
cake, take, Kate, and Tate and many other minimal pairs or sets. 
If such distinctions are not made, language serves as a less effective 
instrument than is otherwise possible. Use of the “correct” (from 
the adult point of view) pronunciation with improved success, 
often enough repeated, would assist in establishing the contrast 
firmly in the speech patterns of the child. However, /k/ would 
finally consist of a very wide range of sounds which will not be 
further divided since the stimulus which led, in this case, to differ¬ 
entiating /t/ and /k/ is missing: there are in English no minimal 

The Phoneme 161 

pairs for any two [k]-like sounds. The process of division therefore 
ceases, and the whole range of [k}-like sounds continues to produce 
the same reaction, that is, to remain as a single phoneme. 

Had Arabic been the language the child was learning, the out¬ 
come would have been different. An Arab child might proceed 
through the same steps and emerge with a /K/ phoneme covering 
roughly the same range of sounds as the English /k/, but the 
process would not come to an end here. Arabic has numerous 
minimal pairs contrasting in two varieties of [k]-like sounds. These 
may be written /k/ and /q/. /k/ is rather far front: /q/ much 
farther back. Such pairs as /kalb/ ‘dog’ :/qalb/ ‘heart’ would 
force the child sooner or later to divide his juvenile /K/ phoneme 
into two ranges /k/ and /q/. Though narrower in their variation 
than English /k/, both these are, like all phonemes, classes of 

An Arab listening to English may identify the consonant of 
English key with his /k/ since it is more or less fronted, and that of 
caw with his /q/ since it is more or less backed. (Actually the 
Arabic /k/ and /q/ are by no means identical with the two varieties 
of English /k/.) This may lead the Arab observer to object to the 
identification of the initial consonants in these words as identical. 
In short, in listening to English, he hears, at least in part, Arabic 
phonemes, not English ones. 

The speaker of Hindi will be unlikely to hear any difference 
between the consonants of key and caw, because his language does 
not force him to establish such a contrast. There is, however, a 
contrast between aspirated stops such as /kh/ and unaspirated 
stops such as /k/. This may be seen from such a pair as/khiil/ 
‘parched grain’ :/kiil/ ‘nail,’ and numerous others. Since the 
initial of key is regularly aspirated, he will equate it with his Hindi 
/kh/, whereas ski will be heard as containing his /k/. He may, 
therefore, object to the assumption that these two sounds are alike. 

12.6 The process of learning a second language involves, 
among other things, learning to make distinctions, both in hearing 
and speaking, that are phonemic in the new language, and learning 
to overlook those distinctions which are not significant, even 
though they may be phonemic in the mother tongue. Often the 
two languages will use very similar sounds but organize them into 
quite different phonemic systems. Learning new uses of old sounds 

162 The Phoneme 

is usually a larger and more difficult part of the problem than the 
mastering of wholly new sounds. Unfortunately, this part of the 
work is easily neglected. Few students of foreign languages are 
aware either of its magnitude or of its importance. Textbooks all 
too frequently aggravate the situation by describing the pronuncia¬ 
tion in a most misleading manner. Thus French /i/ is frequently 
said to be pronounced as is the i in machine , which for most Amer¬ 
icans would be /iy/. It is never so pronounced, though to most 
Americans French /i/ will sound much like their own /iy/. At the 
best, such pronunciation sounds like exceedingly poor and ob¬ 
viously foreign French; at the worst it may be wholly unintelligible. 
Phonemic systems are typically incommensurable. That is, no 
phonemic system can be accurately described in terms of the 
phonemes of another language, and very seldom can even a work¬ 
able first approximation be so stated. 

12.6 The definition of the phoneme in 1.10 will be inadequate 
when the observer cannot hear the language as a native does. To 
make it workable for a non-native it is necessary to add to it some 
objective criteria of the range of sounds that may be included 
within any given phoneme. There are two such criteria, and both 
must be met. (1) The sounds must be phonetically similar. An 
articulatory description such as that sketched in Chapters 2 and 3 
is ordinarily a satisfactory basis for judging similarity — pro¬ 
vided it is made sufficiently comprehensive to cover all the sounds 
concerned and sufficiently detailed to insure that nothing signifi¬ 
cant is overlooked. (2) The sounds must show certain characteristic 
patterns of distribution in the language or dialect under consider¬ 
ation. Two such patterns will be mentioned below. Basically both 
are types of distribution which make it impossible for minimal pairs 
to occur. The latter would of course make the two sounds members 
of different phonemes. Even the occurrence of two sounds in dis¬ 
tributions which would permit minimal pairs (whether or not there 
are any such) will give some functional value to the distinction 
and hence suffice to establish the contrast as phonemic. 

12.7 A phoneme is a class of sounds which: (l) are phonetically 
similar and (2) show certain characteristic patterns of distribution 
in the language or dialect under consideration. Note that this 
definition is restricted in its application to a single language or 
dialect. There is no such thing as a general /p/ phoneme. There 

The Phoneme 


is, however, an English /p/ phoneme. Likewise there is a Hindi 
/p/ phoneme. They are in no sense identical. Each is a feature of 
its own language and not relevant to any other language. 

12.8 The simplest of the patterns of distribution is free varia¬ 
tion. The human vocal apparatus operates with an incredibly high 
degree of precision, but still is far from exact. If the word key is 
pronounced, even by a single speaker, a hundred or so times and all 
the measurable features of each /k/ are measured, it will be found 
that no two are exactly alike. They will, however, cluster about 
certain average characteristics. That is, there will be an average 
duration of closure. Most of the instances will be found to be rather 
near this average, but a few will diverge farther. There will be an 
average degree of aspiration, and most instances will fall near the 
average, though a few will be appreciably more or less aspirate 
that the majority. And so forth. In each of these features there will 
l)e something like the familiar bell-shaped curve of statistics. These 
curves define the range and nature of variation in that one in¬ 
formant’s pronunciation of /k/ in key. 

For the most part the range of such variation is small — so 
small, in fact, that it is not ordinarily discernible except by in¬ 
strumental measurement or by a highly trained phonetician oper¬ 
ating under nearly ideal listening conditions. But occasionally the 
range of variation will be broad enough to be easily heard so that 
it must be given cognizance in phonemic analysis. 

Obviously, any phonetic difference which cannot be consistently 
controlled is of no linguistic significance. Any two sounds which 
are always in free variation cannot be two phonemes, but only two 
points within the range that constitutes one phoneme. 

12.9 In discussing free variation we deliberately selected as 
our example numerous repetitions of precisely the same word. If 
after studying a hundred instances of /k/ in key, we were to repeat 
the experiment on a hundred repetitions of the /k/ in ski we would 
find that the latter also shows variation in every measurable 
characteristic. But the ranges of variation found in the two experi¬ 
ments would not be alike. Some of the characteristics measured 
would differ both in the average values and in the scattering about 
the average. 

Moreover, some of these differences arc quite easily detected by 
cither a moderately trained linguist or, as we have just seen, a 

164 The Phoneme 

wholly unsophisticated speaker of Hindi. The most obvious differ¬ 
ence is in the aspiration. In key, though there is variation in 
aspiration, the range is roughly from moderate to strong; in ski 
the range is roughly from none to weak. There is ample justifica¬ 
tion for considering these two as quite different ranges of variation. 
Why then arc the two not found separating minimal pairs? 

To answer this question it is necessary only to obtain rough 
estimates of the aspiration in a large number of /k/ phonemes in 
many different words. Restricting our attention for simplicity to 
initial position, we will find that /k/ following /s/ is always un¬ 
aspirate or weakly aspirate, while /k/ not preceded by /s/ is 
always moderately aspirate to strongly aspirate. If this is so, 
minimal pairs are obviously impossible. The two [k] therefore 
qualify as members of a single phoneme. 

The distribution just briefly described is known as complemen¬ 
tary distribution. This is the most important, practically at least, 
of the types of distribution which fill the second criterion of our 
present definition of the phoneme. Sounds are said to be in com¬ 
plementary distribution when each occurs in a fixed set of contexts 
in which none of the others occur. In any discussion of phonology 
the only kind of contexts which can be considered are phonologic, 
never morphologic. Thus, English [k“] (unaspirated) and [k h ] 
(aspirated) are in complementary distribution since [k - ] occurs 
in consonant clusters following /s/ as in ski [sk”iy], in medial and 
final clusters before another stop as in act [a;k"t h J, and when not 
initial and preceding a weak stressed vowel as in hiccup Qifk~op]. 
(Note: There is some variation in the details of this pattern from 
dialect to dialect. Some other speakers may have slightly different 
distributions, but will be consistent in following their own pat¬ 
terns.) [k h ] occurs in most other environments, but never in any of 
those listed for [k"]. This is, of course, only a partial statement 
of the whole pattern of variation in /k/; there are also other types 
of differences besides aspiration which must be considered to give 
a full picture. 

12.10 Any sound or subclass of sounds which is in complemen¬ 
tary distribution with another so that the two together constitute 
a single phoneme is called an allophone of that phoneme. A pho¬ 
neme is, therefore, a class of allophones. The ranges of sounds in 
free variation such as were discussed in 12.8 are allophones. 

The Phoneme 165 

12.11 Speakers of the same dialect ordinarily make the same 
phonemic distinctions and the same distribution of the allophones 
within each phoneme. A suggestion as to how a child learns to 
make the necessary phonemic distinctions and only these was 
given in 12.3. This does not serve to explain why two speakers 
have the same distribution of allophones, or even why one speaker 
should exhibit any consistency at all in his use of allophones. 
Without such consistency, complementary distribution would be a 
fiction. Even though that discussion was not integral to the de¬ 
velopment, it imposes the obligation to give some similar sugges¬ 
tions about the origin and significance of allophones. 

In the first place some allophonic distributions are determined 
by physiological factors, or at least the latter make a strong con¬ 
tribution. For example, in English, front allophones of /k/ are 
used near front vowels, as in key, and back allophones of /k/ near 
back vowels as in caw. It would seem reasonable to assume that 
economy of motion might dictate such an arrangement. This is 
probably partially the case, but certainly not wholly so. By com¬ 
parison, in Loma front vowels occur in central (backed) allophones 
after /k g i)/, which seem seldom to be as front as is /k/ in English 
key. For example, /e/, which is normally pronounced much like 
English /i/, is very similar to English /»/ in Loma /ke/. (Note: I 
say “similar,” implying an approximate description.) The other 
two front vowels, /i/ (higher than /e/) and /e/ (rather similar to 
English /e/), have comparable central allophones. In Loma the 
vowels seem to accommodate to the tongue position of the /k g q/; 
in English /kg 13 / accommodate to the tongue position of the 

But economy of motion will not explain all the instances. There 
is no obvious reason to explain why unaspirated allophones of 
/p t k/ occur after /s/. This seems to be very largely just a matter 
of English conventional linguistic habit. It is learned by one 
speaker from another, and has apparently been handed down from 
speaker to speaker through a rather long history. The reason for 
learning it is not that it assists in the use of the language as a tool 
of communication. Its value for this purpose is at best very slight. 
Instead it is a matter of social conformity. If one sounds different 
from his companions he is understood linguistically but suffers 
socially. A person that says key [k“iy] or ski [sk h iy] sounds 

166 The Phoneme 

“funny.” At least this is the case with [sk h iy]. [k“iy] might even 
cause linguistic troubles, as it is too easily confused with /giy/. 

The use of the correct allophones is more important socially 
than it is linguistically. Though obviously of concern to linguists 
for many practical reasons, the allophones stand on the margin 
of his field of study and arc in some respects external to lan¬ 

The use of correct allophones is obviously important to anyone 
learning a foreign language with intent to speak it. To make himself 
understood he must learn to pronounce all the phonemes and to use 
allophones which are sufficiently close to the normal in the language 
to avoid misidentification. Beyond that there is no need, if he is 
merely content to be understood, to worry about the allophones. 
But if he desires his speech to be socially acceptable — that is, 
to sound like that of a native — he must achieve the same use of 
allophones as is normal in the language. Contrary to the experi¬ 
ence of many foreigners, it is quite possible to accomplish this. 
However, it seems that only exceptionally good mimics can attain 
such proficiency without being consciously aware of the problem 
in detail. An understanding of the phoneme principle is, for most 
adults, a sine qua non of successful language learning. It is cer¬ 
tainly the most important single concept that a prospective lan¬ 
guage student can get out of linguistic training. 

12.12 Of the two criteria for the classing of allophones into 
phonemes (12.7) the first, phonetic similarity, can be applied (at 
least in part) without reference to the actual use of the sounds in 
any specific language. Thus, voiced and the corresponding voiceless 
sounds can be said to be phonetically similar, irrespective of the 
language in which they occur. This does not make them the same 
phoneme; both criteria must be met. However, similarity is a 
relative matter. Sounds cannot be said to be either similar or dis¬ 
similar, but only more or less similar. Just how similar two sounds 
must be to fit the definition can only be determined by a consider¬ 
ation of the phonemic system of the language as a whole. If some 
phonemes are found to have both voiced and voiceless allophones, 
then any other pair of voiced and voiceless sounds are certainly 
sufficiently similar to qualify. If, on the other hand, voiced and 
voiceless sounds generally contrast, the chances are that another 
pair of voiced and voiceless sounds are heard as different in the 

The Phoneme 167 

language under consideration. This relativity introduces a certain 
measure of subjectivity into the approach. It is, however, of an 
entirely different kind from the subjectivity which limits the use 
of the method employed in Chapters 2, 3, and 4. 

12.13 The second criterion for classing allophones into pho¬ 
nemes, non-contrastive distribution, has no meaning except in a 
particular language or dialect. It is often susceptible to much more 
objective application than is phonetic similarity. However, this 
criterion poses its own difficulty. The whole distribution of any 
given sound can only be determined by observing the whole lan¬ 
guage. This is obviously impossible. As a practical procedure, we 
must determine distribution within a sample. This always leaves 
the possibility that the examination of additional data may reveal 
additional features of distribution which can compel modifications 
of any earlier conclusions. 

The larger the sample used, the less likelihood there is that more 
material would contradict our earlier formulations. It is, therefore, 
necessary to work with a sample large enough so that the proba¬ 
bilities of error are small. Or, stated another way: with a suffi¬ 
ciently large sample, conclusions can be drawn which have a high 
probability of standing as valid in the light of any additional evi¬ 
dence. It is also necessary that the samples used are representative 
of the speech form under analysis. Finally, once a satisfactory 
sample is obtained, it must be properly used. Nothing can be dis¬ 
regarded, however convenient it would seem to be to do so. 

The techniques of selecting and appraising samples for linguistic 
analysis have received little attention from the point of view of 
the statistical problems involved. Linguists have instead relied 
on the use of very large samples (in most cases) — probably very 
often larger than necessary for their purposes. Most good phonemic 
analyses can be assumed to have appreciable (never absolute, of 
course!) statistical validity. The commonest shortcoming of lin¬ 
guists in this regard is to fail to recognize that their results ulti¬ 
mately rest on data of a statistical nature, and that the conclusions 
must be interpreted with this in mind. 

12.14 In describing the English vowel system from an articu¬ 
latory point of view we found that the nine vowels are describable 
in terms of two significant dimensions, and that in terms of these 
they are arranged in a neat 3 X 3 pattern: 

168 The Phoneme 

i i u 

e o o 

:b a o 

This symmetry affects more than the phonetic base of the pho¬ 
nemes. There are various functional interrelations between the 
phonemes which seem to reflect the same pattern. Thus we can 
consider the phonemes not merely as contrasting entities, but as 
points in a system. 

In some other languages, the phonemic system may show much 
more obvious interrelations. For example, the Turkish vowels 
differ among themselves in three dimensions. They are either rela¬ 
tively front or relatively back, relatively high or relatively low, 
relatively rounded or relatively unrounded. All possible combina¬ 
tions occur, so that the system consists of eight vowels that can 
conveniently be diagrammed as a cube with a vowel at each corner: 

This arrangement can be arrived at from 
a consideration of the articulation, or 
even more clearly from the morpho- 
phonemic relationships. Thus /i u i u/ are 
alike in being relatively high, but in addi¬ 
tion every suffix which contains one of 
these has allomorphs containing each of 
the four. Such affixes are accordingly conveniently described as 
containing high vowels. The allomorphs of such suffixes are con¬ 
ditioned by the vowel of the preceding syllable: /i e/ condition 
/i/; /U 6/ condition /tt/; /i a/ condition /*/; and /u o/ condition 
/u/. This marks off four groups of vowels: front unrounded, 
front rounded, back unrounded, and back rounded. A second 
group of suffixes contains cither /e/ or /a/, both low unrounded 
vowels, /e/ occurs following any front vowel /i ii e 6/, /a/ after 
any back vowel /» u a o/. Thus the arrangement of the Turkish 
vowels in a cubic diagram reflects both structural relationships and 
phonetic character. The Turkish vowels are very evidently units 
in a system, the interrelationships of which are of fundamental 
significance in the language. 

The phonemes of a language are more than individual units to 
be individually identified and described. A third definition of a 
phoneme may therefore be stated as follows: a phoneme is one 

The Phoneme 169 

element in the sound system of a language having a characteristic 
set of interrelationships with each of the other elements in that 
system. These interrelationships are of various kinds. They may 
be reflected in morphophonemic interchanges, in sequences of 
phonemes possible within morphemes, or in the functions of the 
units within stretches of speech. The system may be very sharply 
defined, or more diffuse and indefinite, but such relationships are 
always evident in some degree. 

12.16 These three definitions (in terms of contrasts between 
phonemes, non-contrasting classes of sounds, and systematic rela¬ 
tionships) are complementary. No one of them gives a full picture 
of the nature or significance of the phoneme. Together they pro¬ 
vide an adequate base for either the depth of understanding which 
can be sought in an elementary course or, with only slight modi¬ 
fications of detail, for the most technical discussions. However, all 
three imply some further basic considerations which cannot longer 
be allowed to remain tacitly assumed, but must be discussed if 
these definitions are not to be badly misinterpreted. 

12.16 All our discussion of the phoneme has been on the assump¬ 
tion that the stream of speech is divided into segments, each of 
which can be assigned to some phoneme. There are various ways 
in which that division can be conceived. Most commonly the 
consonants and vowels of English are assumed to follow each other 
in a sequence, each beginning at the end of the preceding one. 
The stresses occur more or less simultaneously with the vowels, 
and the pitches simultaneously with all the others. This model will 
be sufficient for the present, but another possibility will be sug¬ 
gested in Chapter 15. 

Thus, overlooking for the present discussion all stresses, pitches, 
and related phonemes, an English word like bit was assumed to 
consist of a sequence of three sounds. Probably most observers 
would have little difficulty in agreeing with this analysis. But in a 
word like key the practical problem is quite different. Some might 
hear it as a sequence of three items, as the phonemic transcription 
/kiy/ indicates. Others will hear it as only two [ki], still others 
might hear it as four [khiy]. All three of our definitions of the 
phoneme have by-passed or overlooked the problem of how many 
phonemes there are in any stretch of speech, though it is obviously 
basic to each of them. How an utterance is heard and divided 

170 The Phoneme 

(segmented) depends in part on the phonemic background of the 
observer. Different languages divide similar phonetic material into 
different numbers of segments or divide at different places. The 
pattern of segmentation is one of the linguistic patterns of the 
language, and segmentation in accord with those patterns is im¬ 
plied in each of the three descriptions of the phoneme. 

12.17 None of the three definitions makes a clear statement 
on another matter. Speech is a process that involves activity of 
the human brain, articulatory system, and sensory organs, and 
as such can be legitimately and productively studied by psychol¬ 
ogists. The phoneme is not, however, a psychological concept; it 
cannot be defined in psychological terms. Our description of the 
process of learning of a phonemic distinction in this chapter was 
not intended as any sort of definition of a phoneme, but only as a 
subsidiary, or even parenthetical, remark. Obviously, psychological 
processes are involved in the production and recognition of pho¬ 
nemes, but no adequate psychological definition of the phoneme 
has yet been produced. 

Speech also involves certain acoustical phenomena. Somehow, 
these carry the phonemes; but phonemes are not, however, any 
sort of physical reality discernible by instrumental techniques or 
direct observation. The phoneme cannot, therefore, be acoustically 

The phoneme is instead a feature of language structure. That is, 
it is an abstraction from the psychological and acoustical patterns 
which enables a linguist to describe the observed repetitions of 
things that seem to function within the system as identical in spite 
of obvious differences. The phoneme is, in short, a linguistic 
feature only. It is not a feature of a single utterance, but a state¬ 
ment of similarities in numerous utterances. An utterance is not 
properly a sequence of phonemes, but a sequence of concrete ex¬ 
amples of allophones of phonemes. Phonemes are not events but 
classes of events. As such they do not have the same sort of reality 
as a specific part of a specific utterance. In a certain sense they 
are the intellectual creation of the linguist who examines those 
specific parts of specific utterances. But though they are the 
creatures of the observing linguist, he is not free to create as he 
will. The phonemes of a language are a set of abstractions which 
will more adequately describe certain features of the utterances of 

The Phoneme 


that language, past, present, and future, than any other set. They 
are a sort of model of the utterance. The linguist’s task is to find 
that model which most adequately fits the observed facts. The 
language imposes real limitations and often quite narrowly cir¬ 
cumscribes the freedom of the linguist to set up his model. The 
reality of the phoneme lies in these limitations. 




Phonemic Analysis 

13.1 A professional linguist approaching the scientific analysis of 
a hitherto unrecorded language and a layman attempting to attain 
a speaking knowledge of a second tongue face the same funda¬ 
mental problem. Nothing more than crude preliminary work can 
he done until some grasp of the phonemic system is obtained. 
Without this the material is inaccurately perceived, and repro¬ 
duced in terms of the phonemic patterns of the mother tongue, 
or heard and recorded as an unmanageable complex of intergrading 
and unorganized sounds. The layman, if he is “gifted at lan¬ 
guages,” may in time acquire a working control of the phonemic 
system by trial and error. Unfortunately, many never get beyond 
the barest minimum necessary for elementary communication. 
The linguist seeks control of the phonemic system as a conscious 
goal and proceeds toward it by deliberate steps. Some parts of his 
procedure can be of value to anyone who is learning a new lan¬ 
guage, cither as assistance in avoiding common difficulties of 
learners, as a means of speeding his progress, or as a method of 
helping him to understand what he is doing and the reason for 
certain “peculiarities” of the language he is studying. In this 
chapter we will describe some elementary aspects of an analytic 
method which might be used. 

13.2 The first process in analyzing the phonemic system of a 
language other than one’s mother tongue is to make a transcription 


Phonemic Analysis 1 73 

of a sample. A number of utterances are transcribed from the 
speech of a native informant. Great care is taken to insure that the 
transcription is as nearly consistent as possible — that is, that 
every sound is recorded by the same symbol throughout. An at¬ 
tempt is made to record every discernible feature of the speech 
in as great detail as is possible with accuracy. An investigator 
thoroughly trained in phonetics will record a very large number of 
slightly different sounds, using symbols from one of the established 
transcription systems. A person of less preparation for the task 
will record fewer distinctions, and may perhaps need to make up 
ad hoc symbols for many of them. In either case it will be pure 
coincidence if the transcription correctly indicates more than a 
few of the phonemes. The recording will be influenced in some 
degree by the phonemic system of the investigator’s language. A 
competent linguist can suppress this somewhat by reason of his 
wide experience with languages and his training in phonetics; 
his transcription is made not in terms of the phonemic system of his 
mother tongue, but in terms of the cumulative experience of many 
linguists with a large number of languages, and reflects all of the 
distinctions which this polyglot background has taught him to 

Any such preliminary transcription may be expected to depart 
from a purely phonemic representation in various combinations of 
the following: 

(1) It may be over-differentiated. That is, separate symbols 
have been used for two or more variants of a single phoneme, or 
non-linguistic differences may have been recorded. This would be 
the case with an Arab making an analysis of English and writing 
[k] and [q] for /k/. 

(2) It may be under-differentiated. That is, the same symbol 
may have been used for two or more different phonemes, or for a 
combination of allophones belonging to two different phonemes; 
or some linguistically significant feature may not have been 
recorded at all. This might be the case with an American attempt¬ 
ing to analyze Arabic and writing [k] for both /k/ and /q/. 

(3) It may be wrongly segmented. That is, a single symbol may 
have been used for a sequence of phonemes, or a sequence of 
symbols for a single phoneme. This would be the case if an Amer- 

174 Phonemic Analysis 

ican attempting to analyze German were to record [6] for /t§/, or 
a German were to record [t§] for English /$/• 

(4) There may be gross personal errors of various kinds. 

The process of analysis consists of discovering the instances of 
each kind of deviation and applying the necessary correction. By 
this means, the transcription is brought, by successive stages, 
nearer to being an adequate phonemic representation of the lan¬ 
guage. The process may involve a very large number of successive 

13.3 Over-differentiation can be discovered and corrected from 
the record alone by rigorous procedures (provided under-differenti¬ 
ation and gross errors do not seriously obscure the evidence!). It 
is for this reason that a linguist attempts to make the most 
meticulous transcription feasible. This minimizes but does not 
eliminate the incidence of under-differentiation. The record of a 
good phonetician can frequently be analyzed with some case and 
directness. The process is largely one of sorting out of the record 
that information which is significant and disregarding the rest. 
The records of persons not experienced in phonetic transcription 
generally present much greater difficulties, since under-differentia¬ 
tion will very probably occur. No process of sorting relevant from 
irrelevant information can be adequate in such a case, since some 
significant distinctions are wholly omitted from the data. Under- 
differentiation and gross errors can be found and removed only by 
renewed reference to the speech of the informant, though not 
infrequently strong suspicions of their existence and nature can 
be raised from the record. Some of the problems involved will be 
discussed in Chapter 18. Incorrect segmentation can also fre¬ 
quently be detected from a good record, particularly such a 
transcription as an experienced field linguist is likely to make. 
Being aware of the dangers, he may take extra care to record 
adequate data at critical points. 

13.4 For the present we will overlook the possibilities of under- 
differentiation and gross errors and assume that we have a 
phonetic transcription which deviates from phonemic only in over- 
differentiation, and later in the chapter also in incorrect segmenta¬ 
tion. Of course, such an assumption can be made for purposes of 
presentation of method only. As with morphologic analysis, dis¬ 
cussed in Chapters 6 and 7, all the various processes must be 

Phonemic Analysis 175 

carried out more or less simultaneously. Indeed, it is usually con¬ 
venient and necessary to conduct grammatical and phonologic 
analysis together. This is done only as a matter of practical 
procedure; the two are not to be confused. 

Before the analysis of a language is completed the analyst must 
assure himself that the phonologic results are independent of the 
grammar. That is, he must be able to describe the phonemic sys¬ 
tem and support his analysis without any appeal to any morpho¬ 
logic results beyond the bare assertion that pairs of utterances are 
actually different. The converse is emphatically not true; an 
adequate grammatical description must presume an adequate and 
complete phonologic description. The morphology must be stated 
in terms of morphemes, which in turn are stated in terms of the 
phonemic system of the language. No other basis is wholly ade¬ 
quate or satisfactory for anything beyond preliminary statements. 
The most frequent limitation on further progress in understanding 
the grammar of a language is an inadequate phonemic analysis, 
commonly one which has made an adequate analysis only of the 
consonants and vowels. 

13.5 In the last chapter we discussed three complementary 
definitions of a phoneme. Each of the three has implications for 
phonemic analysis. The method advanced in Chapters 2, 3, and 4 
was based on one of these definitions alone, and was in that 
measure inadequate. For the present purpose, the most important 
of the three is that presented in 12.7, but both the others con¬ 
tribute in important ways. A phoneme was defined as a class of 
sounds and two criteria were set up for the assignment of any two 
sounds to a single phoneme. The first of these was phonetic 
similarity. This imposes the requirement that the analyst take care 
not merely in transcribing the sounds of the language so that cacli 
symbol in his preliminary notes will always represent the same 
sound, but also that he record with care the phonetic nature of each 
sound. This can be done by using symbols from a conventionally 
established set in which each symbol carries a phonetic definition, 
or by carefully describing the phonetic character of each sound 
represented by an ad hoc symbol or by an unorthodox use of a 
familiar symbol. In this chapter and the accompanying problems 
we will use symbols in values defined in Chapter 14. 

The second requirement of our definition of a phoneme is certain 

176 Phonemic Analysis 

patterns of distribution: complementary distribution or free varia¬ 
tion. The demonstration of either of these can be made step by 
step, but the procedures are somewhat different. We will first 
describe a method by which phonemic groupings based on com¬ 
plementary distribution can be detected. This consists of three 
steps: finding suspicious pairs, framing a hypothesis, and testing. 

13.6 First, tabulate all the sounds found in the sample, 
using suitable phonetic classifications. List suspicious pairs, that 
is, pairs of sounds which seem to be phonetically similar, and hence 
possibly allophones of the same phoneme. One sound may be in¬ 
cluded in several such pairs. Disregard pairs of sounds which are 
so different that they cannot be allophones of the same phoneme. 
For example, if the data shows [m k k h J, you may assume that 
M and [k] will be assigned to separate phonemes, but [k] and 
[k h ] may be allophones of one phoneme. The latter, therefore, are 
to be considered as a suspicious pair. 

The following are commonly allophones of the same phoneme, 
and hence should be suggestive of sounds which should be listed 
as suspicious pairs. Examples are given in [ ]. 

Corresponding voiced and voiceless sounds [k g] £s z] 
Corresponding stops and fricatives [k x] 

Bilabial and labiodental stops, fricatives, nasals, or stops and 
fricatives [b b] [v v] [m m] [p f] 

Dental, alveolar, and retroflex stops, fricatives, laterals, or 
stops and fricatives [t t] [t t] [d 1 ] [t 6 ] 

Alveolar and alveopalatal fricatives [s §] 

Palatal and velar stops or fricatives [k k] [k k] [x x] 
Corresponding aspirate and unaspirate stops [t h t] 

All nasals, except that [m] is usually distinct [n 13 ] [n fij 
All varieties of [r]-like sounds and many [l]-like [r f] [F r] [f 1] 
Dental and alveolar flaps and stops [d r] [t f] 

Uvular [r] with velar or uvular fricatives [v r] 

[h] and all unvoiced velar or palatal fricatives [h x] 

All pairs of adjacent vowels [i u] [i i] [ee a] [0 0 ] 

Semivowels and labial, palatal, or velar fricatives [w u] 

Such a list cannot be exhaustive. There are other combinations of 
sounds that occasionally will be found to be allophones of the 
same phoneme. 

Phonemic Analysis 1 77 

13.7 Examine the distribution of each member of each sus¬ 
picious pair. Attempt to frame a hypothesis which will account 
for their distribution. This hypothesis should be based either on 
observed peculiarities, or on parallelism with the situation already 
found in similar pairs. For example, if you have established that 
[pj and [b] are in complementary distribution and are now con¬ 
sidering [t] and [d], it is reasonable to try a hypothesis similar to 
that found to fit in the case of [p] and [b]. There is no assurance 
that they will show close parallels, but since phonemes are elements 
in more or less integrated systems, there is an appreciable prob¬ 
ability that they will do so. 

The distribution of allophones may be conditioned by any 
phonologic feature in the language. Among the commoner are the 

Immediately preceding or following phonemes. 

Preceding or following phonemes at a somewhat greater dis¬ 
tance, though the probability decreases with distance. Vowels 
of successive syllables not infrequently affect each other. 

Position in syllable, word, clause, etc., provided these can be 
phonologically defined. 

Relationship to stress, pitch, or other similar features. 

Combinations of two or more of these factors. 

The conditioning operating in complementary distribution often 
parallels assimilation or other well-known morphophonemic 
changes. Italian /n/ has allophones [n] and [q], the latter occur¬ 
ring for example in /bianko/ ‘white’ [biaqko]. This distribution 
parallels the English assimilation of /n/ to /q/ before /k g/. There 
is, however, this basic difference: In English two different pho¬ 
nemes are involved, since /n/ and /q/ contrast elsewhere as in 
sun /s5n/ : sung /sSq/. In Italian very similar sounds are two 
allophones since [n] never contrasts with [q]. Many such cases 
can be cited. The types of sequences which arise as the result of 
morphophonemic changes should be looked for in framing a 
hypothesis concerning any suspicious pair of sounds. Incidentally, 
if the hypothesis does not prove correct it may suggest something 
about morphophonemic relationships which should be followed up 
in a different place in the analysis. 

In some cases the conditioning may seem to be “inexplicable.” 

178 Phonemic Analysis 

Actually, it does not matter whether an “explanation” can be 
made or not, for any such “explanation” is nothing more than 
another description of an observed distribution. Many types of 
correlations occur for which it has never been felt necessary to 
coin a label. This lack of a special term does not, of course, affect 
their validity in any way. 

13.8 After a hypothesis has been framed, test it by tabu¬ 
lating the distribution of each sound in relationship to the 
suggested conditioning factor or factors. Be sure in doing so to 
tabulate all the relevant data. It is at this point that care must 
be taken to ensure the statistical validity of the result. The 
hypothesis must account for every occurrence of each member of 
the suspicious pair. If the tabulation shows a correlation between 
the distribution of the sounds under test and the hypothetical 
conditioning factor, accept the hypothesis as a working basis. 
Consider the conclusion as very tentative unless the number of 
instances included in the tabulation is appreciable. Obviously, if 
the sample includes only one occurrence of each of two sounds, the 
two can almost always be shown to be in complementary distribu¬ 
tion unless the examples are a minimal pair. The more complex the 
hypothesis, the more extensive the data required to support it. 

If the hypothesis does not work out, it must be either rejected 
or modified. When a tabulation shows quite close correlation but 
a few exceptions, the exceptions should be carefully examined. 
They may seem to have some common factor which may be taken 
into account in a modification of the original hypothesis. With a 
new hypothesis, a new tabulation must be made. When there seems 
to be no further reasonable hypothesis available, consider that the 
two sounds are members of separate phonemes. However, in such 
a case bear in mind that there may be complementary distribu¬ 
tion, with a conditioning factor that has not yet been noticed. 
Your conclusions are all tentative until the analysis is completed 
and they have been checked one against the other. 

13.9 We may illustrate the method by a partial analysis of the 
following Spanish words. The amount of data, here as in most of 
the workbook problems, is very close to the minimum. In field 
work many times thirty words would be considered essential. 
Moreover, these words were selected to illustrate certain parts of 
the system only. Several phonemes are not represented at all. 

Phonemic Analysis 179 







[bay a] 













we give 






to owe 





to endure 



to earn 







[ponderoso] heavy 

to enjoy 


I put 






I have 










The first step is to tabulate the sounds recorded and to identify 
the suspicious pairs. The latter have been indicated by loops. 

Voiceless unaspirate stops /p\ ft\ 

Voiced unaspirate stops 

Voiced fricatives 

Voiceless groove fricative 


Flap and trill 

Nasals ra (j T 

The first pair to examine is [p] and [b]. Both occur initially, 
so no hypothesis based on position in the word is feasible. It could 
be a matter of the following phonemes, so we make the following 

Before [i] Before [e] Before [a] Before [o] Before [u] 
[p] / // / // 

M / // / / 

This table certainly shows no evidence of complementary distribu¬ 
tion. We might set up the hypothesis that [p] and [b3 are condi¬ 
tioned by the next consonant. If we tabulate this we get the fol¬ 
lowing result: 

Before 1 ysSPrpnq- 

[b] / / / / / 

[p] ////// 

180 Phonemic Analysis 

Both occur before [f], so this hypothesis is unworkable. Even if 
we did not have [pero] and [bufo], the tabulation would hardly 
prove anything, since, with so many different positions being 
tabulated, the data is insufficient. We must conclude that [p] and 
[b] are not in complementary distribution, or that our present data 
is insufficient to indicate the pattern. 

The second pair to be examined is [b] and [u]. It is immediately 
noticeable from the word list that [v] does not occur initially. We 
therefore set up the hypothesis that [b] is found only initially in 
words, [u] only medially. We then check this hypothesis by the 

following tabulation: 







The hypothesis is sustained, and we conclude that [b] and [V] are 
ollophoncs of one phoneme, which we will write /b/. We may now 
replace the transcription [auana] by [abana]. This is still written 
in [ ] rather than / / because we have not yet established the 
status of the other sounds. 

On the assumption that [d] and [6] parallel [b] and [V] we 
next turn to these two and test the equivalent hypothesis by the 

following tabulation: 








This hypothesis is not sustained. Nevertheless, it seems significant 
that [$] is found only in medial position, so we examine the 
instances of medial [d]. These suggest a modification of the 
hypothesis — that [d] occurs initially and after [n]. We therefore 
make a new tabulation: 


After [n] 

After vowels 




This hypothesis is sustained. The two are allophones of one pho¬ 
neme which we will write /d/. [naba] may be rewritten [nada]. 

Phonemic Analysis 181 

This should raise the question as to why a similar distribution was 
not found for /b/. The reason was simply lack of data. More ex¬ 
tensive lists would certainly include some such word as [bomba] 

The pair [g] and [Y] might be expected to show the same rela¬ 
tionship. Make the necessary tabulation to test this and the re¬ 
maining untested pairs. 

13.10 While examining the data to detect cases of comple¬ 
mentary distribution which will establish two sounds as allophones 
of a single phoneme, notice should be taken of any evidence which 
will prove the converse. The most valuable is, of course, minimal 
pairs. You must take care to insure that the pairs are indeed 
minimal, rather than differing by some feature which has not been 
adequately recorded. However, minimal pairs may be very hard 
to find, particularly in the small corpus of material that is likely 
to be available at the early stages of an investigation. You know 
from experience with English, in which you can draw from a very 
large vocabulary, that minimal pairs are sometimes quite difficult 
to find. Actually, English seems to have more minimal pairs than 
many languages. In some they are exceedingly infrequent. Minimal 
pairs are useful when found, but not necessarily to be expected, 
and not essential to the work of analysis. 

In 2.23 another method was suggested. It involves the use of 
sub-minimal pairs, that is, items that differ in only two, or per¬ 
haps three, respects. These are of value because they restrict 
severely the hypotheses that can be suggested as conditioning any 
possible complementary distribution. When the number of possible 
hypotheses is small, it is possible to examine each of them. If all 
can be ruled out, then it may be taken as proved that the two 
sounds are not allophones of one phoneme. Minimal pairs arc 
merely the limiting case, where there is no possible hypothesis of 

13.11 Free variation presents somewhat different problems. 
When two sounds are in free variation, they can obviously be found 
in similar environments — and, if the data is adequate, in identical 
environments. That is, they will occur in what might at first sight 
be taken for minimal pairs. These pairs are the same morpheme 
or sequence of morphemes occurring sometimes with one variant 
and at other times with another variant of the phoneme under 

182 Phonemic Analysis 

consideration. They are not properly minimal pairs, because by 
definition minimal pairs must differ in both content and expression. 

To put it in another way, given two utterances which are alike 
except in one feature, this may constitute evidence that the two 
sounds are phonemically distinct, or that they are variants of one 
phoneme. Which conclusion is to be drawn depends on whether or 
not the two utterances differ in content, and if they do, whether 
or not the difference of content is consistently correlated with the 
difference in expression. If there is no correlation of differences in 
content and expression, then we have free variation. 

This, however, is not enough. For many Americans both [wiO] 
and [wi$] occur with no discernible conditioning. This is therefore 
a case of free variation. However, very many words always have 
[0], and many others always have Indeed, minimal pairs can 
be found to show that [0] and [8] are phonemically distinct. The 
case of [wiO] and [wi'S] is an example of free variation between two 
alloraorphs of a single morpheme and is not phonologically rel¬ 
evant. To establish that two sounds are members of a single pho¬ 
neme, free variation must be observed throughout a phonologically 
defined range of occurrences, not merely in certain morphemes. 
This means that to use this criterion in phonemic analysis it is 
necessary to have a large number of items, each of which is re¬ 
corded several times. It is not likely that such evidence will be 
obtained in a preliminary transcription. It is usually necessary to 
return to the informant to elicit additional material to either 
support or refute any hypothesis of free variation. 

13.12 The examination of each suspicious pair and a decision 
in each case either for or against the hypothesis that they belong 
to the same phoneme is not enough. Phonemes are not isolated 
entities. Our third characterization of the phoneme in 12.14 was 
as an element in a more or less integrated sound system. It is, 
therefore, essential that the individual conclusions be weighed 
against the background of the total phonemic system as it emerges 
in the course of the work. 

Moreover, piecemeal analysis may easily lead to inconsistencies. 
In English Qj”] and Q> h 3 can be shown to meet the requirements 
for inclusion in one phoneme. The same technique will produce 
the same result with [p“] and [b]. This might seem to indicate 
that [p“], [p h ], and [b] should all be classed together. However, 

Phonemic Analysis 1 83 

[p h ] and [b] are easily proved to belong to separate phonemes. 
[p“] must be assigned cither to /p/ or to /b/; it cannot belong to 
both. Nor can [p“] be set up as a third separate labial stop pho¬ 
neme, since it does not contrast with either of the other two. A 
choice must be made as to which of the two it will be assigned to. 

Traditionally, [p“] is assigned to /p/. There is some evidence 
to support this, but it is complex and somewhat tenuous. Any 
analysis not based on full and detailed information about the dis¬ 
tribution — information far beyond what would be found in a first 
investigation — will provide no basis for a decision. The assign¬ 
ment of [p"] would then have to be arbitrary. Decisions which arc 
wholly or very largely arbitrary are frequently required. 

If arbitrary decisions are to be made, they frequently must be 
made in several places within one phonemic analysis. For example, 
the assignment of [t“] to /t/ or /d/ and of [k~] to /k/ or /g/ rests 
on indecisive evidence parallel to that bearing on [p“]. To 
make the systematic relationships within the phonemic system 
evident, all three of these decisions should be made the same 
way. Thus if [p“] is assigned to /p/, then [t“] should be assigned 
to /t/ and [k“] to /k/. 

13.13 So far we have assumed that our preliminary record is 
correctly segmented. This is not necessarily the case and cannot 
be taken for granted. However, there is no need of raising a ques¬ 
tion about every segmentation point; certain types of sounds and 
sound sequences are particularly likely to be differently treated in 
different languages. These should be considered as suspicious and 
checked in the analysis. 

The following are most likely to be recorded as sequences while 
requiring analysis as single segments: 

Stop 4- aspiration (aspirate stop) 

Stop + homorganic fricative (affricate) 

Alveolar or dental stop + lateral 

(lateral affricate) 

Homorganic nasal -f stop 
Glottal stop + stop (glottalizcd stop) 

Consonant + semivowel 
Stressed vowel + vowel glide or semivowel 
Semivowel or vowel glide + stressed vowel 

[th] [kh] [ph], etc. 
M [tO] [kx] 

m m 

[mb] [nt] [qk] 
[’k] [k?] 

[kw] [ny] [ly] 
[ow] [iy] [e*] 

[ye] Du] [wu] 

184 Phonemic Analysis 

13.14 Such sequences may be interpreted as single segments if 
they are limited in number and: 

(1) If they are in complementary distribution or free variation 
with single sounds. For example, in Kikuyu (Kenya) [mb] and [b] 
are not phonemically different. There are two options; either [mb] 
is an allophone of /b/ (a unit phoneme), or [b] is a variant of some 
such sequence of phonemes as /mb/. In Kikuyu the preferred 
interpretation is that [b] and [mb] are allophones of a phoneme 
/b/, and that hence a segmentation between [m] and [b] is 

(2) If the sequences are such that treatment as unit phonemes 
will result in an overall simplification of distributional statements. 
For example, in Bannock (Idaho) the following simple consonants 
are found to be phonemic: /ptkbdgmnij wsh’/. In addi¬ 
tion there are the following phonetic clusters: [ts t§ kw dz d i gw]. 
These are all of suspicious types, so that it would be quite feasible 
to interpret them as unit phonemes: /c 6 k w j ] g*/- This has the 
advantage of eliminating all clusters and thus simplifying a de¬ 
scription of phoneme sequences. 

(3) If the clusters seem to occupy a place in the total phonemic 
system of the language comparable to that of undisputed unit 
phonemes. For example, in Hindi, [6 £h ) jh] parallel the four 
series of stops in many respects and fill out the phonemic system 
to a high degree of symmetry. 

Naturally, any such criteria must be used with caution, but not 
infrequently it will be found that two or all three of them reinforce 
each other and point very clearly to treatment of certain sequences 
as unit phonemes. 

13.16 Conversely, what the investigator perceives as single 
segments may be better treated as sequences under the following 

(1) If apparent single sounds are in complementary distribution, 
or otherwise not phonemically distinct, from sequences of sounds. 
This is the exact converse of the case cited above. For example, in 
English a word like bottle is pronounced either as [b&tQ with a 
syllabic [|] or as [batil] with a short but distinct vowel. These are 
in free variation in all comparable environments, and so not 
phonemically distinct. We have considered /b&til/ as the better 

Phonemic Analysis 185 

interpretation, making the single segment Q] a variant of the 
sequence of phonemes /»!/. 

(2) If what is recorded as a single sound occurs in a distribution 
which is typically occupied by a sequence of phonemes. For 
example, in Egyptian Arabic, vowels are phonemically either long 
or short, as may be seen in the contrast /ba-rid/ ‘cold' : /bari-d/ 
‘mail.’ Single short consonants occur after cither long or short 
vowels. Clusters of two consonants occur commonly only after 
short vowels. There are no clusters of three or more consonants. 
Long consonants occur commonly only after short vowels. There 
are no clusters of a long consonant and any other consonant. That 
is, long consonants have the same restrictions on occurrence as do 
clusters of two consonants. They may, therefore, be interpreted as 
clusters of two identical consonants: [bar ad] ‘cool’ is interpreted 
as /barrad/, in contrast with /barad/ Tile.’ This brings the pattern 
into line with such words as /zikrin/ ‘mention’ or /kursi/ ‘chair.’ 

(3) If such an interpretation will fill out gaps in a list of clusters. 
For example, in Swahili there is an extensive series of clusters of 
the type /Cw/, but no [bw]. There is, however, a single segment 
[u] which can be interpreted as /bw/. 

The following types of sounds are often best interpreted as 
sequences of phonemes: 

Long vowels 
Long consonants 
Syllabic consonants 
Nasal vowels 

Affricates and other types mentioned 

[a-] =/aa/ 

[m-] = /mm/ 

M =/Vn/~/n V/ 
[a] “ /an/ ~ /aij/ 

in 13.13 [fi] = /tS/ 

13.16 In some languages the pitch phonemes are distributed in 
a way that correlates with vowel length. For example, in Loma a 
short vowel has either a low pitch /'/ or a high pitch /'/. Long 
vowels and diphthongs have low, high, rising, or falling pitch. 
The most useful way of interpreting this is to assume that the 
long vowels and diphthongs are sequences of two short vowels, 
each of which has an associated pitch phoneme. Thus [a-] with 
high pitch is fk a/, with falling /a&/, and with rising /&a/- This is 
a very special instance of cogent evidence for resegmenting. Similar 
patterns do not occur in all languages. Some have only a single 

186 Phonemic Analysis 

pitch phoneme with any vowel, long or short. Others have one, 
two, or even three pitches associated with a single vowel irrespec¬ 
tive of length. 

13.17 Throughout the preceding discussion we have seemed 
to be inordinately concerned with transcription. The process of 
phonemic analysis has been presented as a matter of starting from 
a non-phonemic transcription, juggling symbols, and gradually 
approximating a phonemic spelling. This is the easiest way to 
describe the process, and actually it is very nearly what is done 
by a practicing linguist. However, the linguist’s concern is not 
with the symbols as such, but with the classes of sounds behind 
the symbols. His task is to start with a raw impressionistic inter¬ 
pretation of sounds heard and gradually discern interrelationships 
between the sounds, until he is finally able to make simple and 
general statements about the sound system of the language. His 
objective is the most general and the simplest description which 
is adequate to describe the totality of linguistically significant 
features. Through the process more is done than merely to juggle 
the symbols, which are merely labels for sounds or classes of 
sounds; the basic nature of the symbolization is altered. At the 
start, the symbols stand for occurrences of sounds which are im¬ 
pressionistically identical, having the same acoustic effect as far 
as t he observer can judge. At the end, the relationship between the 
symbols and the sounds is much less direct. For a direct relation 
between sound and symbol has been substituted a direct relation 
between structure and symbol. A phonemic symbol such as /b/ 
does not stand directly for any phonetic entity, but for a structure 
point in English phonology. That this structure point is generally 
associated with an acoustic event which may be described as a 
voiced bilabial stop is important, of course, but it is not the 
primary significance of /b/. The manipulation of the transcription 
is merely the outward expression of an analytic process which 
seeks to penetrate behind the acoustic or phonetic facts to the 
linguistic structure. 


c < < -c-c-c %'<■<: <v<vvnv<v 


Articulatory Phonetics 

14.1 The science of linguistics depends on various other disci¬ 
plines for certain basic concepts and methods. One of the most im¬ 
portant contributions is made by phonetics, particularly the older 
branch known as articulatory phonetics. This branch is concerned 
with the study of sounds usable in speech in terms of the mechan¬ 
isms of their production by the human vocal apparatus. It has 
provided linguists with the greater part of their technique and 
terminology in handling sounds. 

Phoneticians at one time set as their goal the exact and detailed 
description of every sound. As work proceeded, it soon became 
evident that this objective could never be reached. The human 
vocal apparatus can produce an infinity of sounds. The only limit 
on the number which can be identified is the instruments used. 
Nor can the field be feasibly restricted by excluding some on the 
basis that they do not figure in speech. Certainly some are rarely 
so used, but some languages use sounds which from our ethno¬ 
centric viewpoint seem very odd. A linguist must be prepared to 
find any vocal sound in use in speech. General phonetics has 
accordingly come to seek after a comprehensive description of 
classes of sounds and of the general mechanisms of speech produc¬ 
tion. One objective is to provide a means by which any given sound 
can be classified and described with whatever degree of precision 
may be necessary. This provides the linguist with the tools needed 
to handle the phonetic systems which he finds in his work. 


188 Articulatory Phonetics 

14.2 Speech production is an incidental activity of the respira¬ 
tory system. Most of the time, air passes into and out of the lungs 
more or less silently; only when there is some obstruction is 
appreciable sound produced. Speech further requires that there 
be easy and effective control of these sound-producing obstructions. 
This limits the mechanisms of interest to linguists largely to move¬ 
ments within the mouth, pharynx, and larynx. While these have 
always received the greater attention from phoneticians and 
linguists, it must not be forgotten that the motive power for sound 
production arises largely in the activity of the thorax, which there¬ 
fore has an important, though quite different, effect on speech 

The traditional phonetic classification of speech sounds is based 
primarily on three variables, each of which will receive separate 
discussion. These are: (1) the activity, if any, in the larynx, most 
familiarly thought of in terms of the dichotomy between voiced 
and voiceless sounds; (2) the place of maximum constriction in 
the mouth or pharynx, usually referred to as the point of articula¬ 
tion; (3) the type of sound-producing or sound-modifying mechan¬ 
ism in the mouth or pharynx, often referred to as the manner of 
articulation. While these three provide a basic classification, alone 
they are inadequate to specify sounds with sufficient precision for 
all linguistic work. Not infrequently it is necessary to mention 
some secondary articulation, that is, some other feature or features 
which can be conceived as modifications imposed on a basic speech 
sound defined in terms of the three classical phonetic variables. 

14.3 The larynx is a cartilaginous structure at the summit of 
the trachea. Its chief importance in speech is that it contains the 
vocal cords, which are two horizontal folds of elastic tissue, one 
on either side of the passage. They may be opened so as to cause no 
obstruction (as in normal breathing), completely closed, or par¬ 
tially closed so as to produce various sorts of audible sounds. Since 
they arc elastic they may be caused to vibrate if brought together 
while air is forced between them. The resulting sound is voice. 
Though voice may be produced by other elastic organs, voice pro¬ 
duced by the cords is by all means the most important type and is 
usually referred to as “voice” without qualification. Voice is char¬ 
acterized by a definite pitch, which is controlled, in large part, by 
adjustment of the tension on the cords. Many speech sounds are 

Articulatory Phonetics 189 

basically glottal voice modified in various ways by the shapes of 
the respiratory passage above the larynx. 

The passage of air through a narrowly constricted opening pro¬ 
duces a second fundamental type of sound known as friction. 
Friction differs from voice in that it has no definite pitch. If the 
vocal cords are incompletely closed, glottal friction may be pro¬ 
duced. This substitutes for voice in whisper. The exact quality 
of the whisper is determined by the shape of the passages above 
in ways that parallel closety the control of quality of voiced sounds. 
No glottal sound can escape modification by the position of the 
vocal organs in the mouth and pharynx. 

Nevertheless, it is customary to list in an inventory of sound 
types at least three glottal sounds, each of which is provided with 
a phonetic symbol — [h fi ’3- They may be taken as cover sym¬ 
bols, useful primarily in situations where the articulatory modifica¬ 
tion does not particularly need to be noticed. 

The glottal stop [*] is produced by the closure, opening, or 
closure and opening of the vocal cords. It is, of course, not possible 
to have a voiced glottal stop similar in mode of formation to 
another voiced stop. Glottal stop frequently accompanies various 
articulations, producing glottalized sounds. Most common are glot¬ 
talized stops, but glottalized fricatives and resonants also occur. 

Customarily, [h] and [fi] are listed as "glottal fricatives,” voice¬ 
less and voiced respectively. For [h] this description is generally 
adequate, since the mechanism of formation is quite comparable 
to that of [f 0 s x], etc. But for [fi] it is nothing more than a con¬ 
venient label, since a mechanism such as that for [vSzV] is 
obviously impossible. A rapid change in the pitch of the voice 
seems to be involved in [fi], as the cords are relaxed or tightened 
in passing from voice to an [h]-like position, or the reverse. The 
pitch change is so rapid that it is not heard as such, but produces 
an acoustic effect very similar to [h]. 

Since during the pronunciation of [h] or [fi], the mouth may be 
in position for almost any sound, these two may be considered as 
existing in a number of varieties, each of which can be alternatively 
considered as a variety of some vowel or other speech sound 
defined by that articulation. 

Other adjustments of vocal cords produce falsetto, ventrilo- 
quistic voice, and various other phenomena of lesser importance. 

190 Articulatory Phonetics 

The mechanism of most of these is poorly understood, and they 
are only rarely of linguistic importance. The diagrams illustrate 

Open Closed Voice Whisper 

The Vocal Cords from Above and Behind 

the appearance of the cords as seen from above in four of their 
most important positions. 

14.4 The passages may be partially or wholly obstructed or 
altered in shape by various organs known as articulators. Often 

the obstruction is 
formed by two organs, 
one of which is mov¬ 
able and approaches 
the other, which is 

The roof of the 
mouth is divisible into 
four portions. Just be¬ 
hind the front teeth is 
the tooth ridge or 
alveolae, the portion 
of the roof of the 
mouth which is con¬ 
vex in shape. Behind 
the alveolae is the 
palate (or hard pal¬ 
ate), which is a plate 
of bone thinly covered 
with other tissue. It is immovable. Behind the palate is the 
velum (or soft palate), which is muscular and movable and may 
be raised to close off the nasal passage. This is called velic closure 
in contrast with velar closure, which is the closure of the oral 

Articulatory Phonetics 191 

passage by the tongue against the lower surface of the velum. 
The uvula is a small, flexible appendage hanging down from the 
posterior edge of the velum. 

Around the alveolae and palate are the teeth. In many articula¬ 
tions the edges of the tongue lie against the molars closing the oral 
passage at the sides. This; 
however, is hot independently 
variable, but seems to be 
largely incidental'to the height 
of the tongue,' so that the 
molars are of little signifi¬ 
cance in phonetics. The term 
teeth, when unqualified, refers 
to the upper front teeth only. 

The lower teeth are much less 
often significant, but do play 
some part in speech forma¬ 

The tongue is conveniently 
divided into four portions. 

The apex is the portion that 
lies at rest opposite the alveo¬ 
lae; it may articulate against the teeth, alveolae, or palate. The 
front is the portion that lies at rest opposite the fore part of the 
palate. It may articulate against the alveolae, palate, or velum. 
The back or dorsum is the portion which lies at rest opposite the 
velum or the back part of the palate; it may articulate against 
the posterior part of the palate, or any part of the velum, or the 
uvula. The root of the tongue forms the front wall of the pharynx. 
While not often listed as an articulator, this portion of the tongue 
contributes to sound formation in altering the size and shape of 
the pharynx. 

Both the lips are movable and of importance in speech. How¬ 
ever, the lower lip is more flexible and more variously used. There¬ 
fore, when used without qualification, the term lip refers to the 
lower. It may articulate against the upper lip or the teeth. The 
spreading and rounding of the lips, and their protrusion, are 
among the most important secondary articulations. The lips have 
the advantage of being visible and hence should be visually ob- 

192 Articulatory Phonetics 

served, as should any other feature which can be seen. One of the 
commonest failures of students in phonetic work is to neglect the 
visual evidence of sound formation. 

14.5 The articulators taken in pairs serve to define the basic 
points of articulation. Intermediate points of articulation are 
common, but the following terms do, however, serve to classify 
sounds with sufficient precision for most linguistic purposes: 

Lower Articulator 

Upper Articulator 



(lower) lip 

upper lip 


(lower) lip 

(upper) teeth 



apex of tongue 

(upper) teeth 


apex of tongue 



apex of tongue 
(see below) 




front of tongue 

alveolae and far front 
of palate 


front of tongue 

front of palate 



back of tongue 

back of palate 


back of tongue 



back of tongue 

extreme back of 
velum or uvula 

Extreme forms 

of dental sounds in which the apex of the tongue 

actually protrudes beyond the teeth are sometimes distinguished 
as interdental. 

In retroflex articulation the tip of the tongue is turned back 
so that the closure is relatively far back on the palate. Because of 
this turning back, the closure may be made with the underside of 
the tip. 

Alveopalatal sounds are, as the term indicates, intermediate 
between alveolar and palatal. Both the alveolae (or perhaps only 
the back portion) and the front part of the palate are involved. 
The apex of the tongue is not involved in alveopalatal or palatal 
sounds and should be kept down. Close attention to this will help 
avoid one of the common errors of Americans with some of these 

Articulatory Phonetics 193 

sounds — that is, the tendency to substitute an alveolar sound 
followed by [y]. 

In uvular sounds the articulation is as far back as possible. 
American English velars vary from palatal to true velar. Pre¬ 
palatal and uvular sounds may be considered as just beyond the 
limits of variation of velars in normal American speech. 

In labiovelar stops there is a simultaneous closure at two points, 
bilabial and velar. These sounds are often indicated by digraphs 
[kp] and [gb]. They are not, however, [k] followed by [p], or 
[g] followed by [b], but in a sense these two stops said simul¬ 

14.6 There are three basic general sound types: stops, frica¬ 
tives, and resonants. The vast majority of speech sounds fall into 
one of these categories. 

Any sound is modified by the shape of the air passages and 
cavities open to it. Thus, if voice is produced by the vocal cords, 
the sound will be quite different according as the oral passage, the 
nasal passage, or both are open to it, and according to the shape of 
the oral passage. (The shape of the nasal passage is not subject to 
controllable variation.) Thus a wide variety of voiced resonants are 
produced. Resonants are sounds in which the only function of the 
mouth and nose is to modify the sound already produced in the 
larynx. That is, there is no constriction in the mouth producing 
friction or other appreciable sound. Voiced resonants are generally 
more common than voiceless resonants. In the latter there must 
be some sound other than voice produced in or near the larynx — 
generally weak friction at the partially constricted vocal cords. 
Voiceless resonants are frequently not phonemically distinct from 
one another, and hence are commonly classed together as a 
glottal fricative [h]. 

If there is a constriction producing friction anywhere in the 
mouth, the resulting sound is a fricative. There may be simul¬ 
taneous voice in the cords, in which case the sound is a voiced 
fricative, or the cords may be inactive, in which case the sound is 
a voiceless fricative. 

Stops are produced by complete closure. Prolonged closure does 
not, of course, produce sound. It is only the act of closing or 
opening, or both, or glottal activity during closure (which cannot 
be long continued) that produces the sound known as stops. They 

194 Articulatory Phonetics 

therefore differ fundamentally from resonants and fricatives in 
that they cannot be indefinitely prolonged. 

14.7 Fricatives vary not only according to the position of the 
constriction (point of articulation), but also according to its shape. 
Three kinds are shown in the chart, page 200. In slit fricatives the 
opening is relatively wide horizontally and shallow vertically. The 
tongue (if it is the organ concerned) is relatively fiat. In groove 
fricatives the tongue is more or less grooved by a raising of the 
edges. The result is that the opening is much narrower horizontally, 
but may be deeper vertically than in slit fricatives. Groove frica¬ 
tives all have more or less of an [s]-like quality, and are for this 
reason sometimes called sibilants. In both of these types the open¬ 
ing is median, that is, it includes the mid line of the mouth. A 

Dental slit frica- Alveolar groove frica- Alveolar lateral frica¬ 
tive tive tive 

[0] 03 C«] 

lateral fricative has the opening on one side of the mouth only, 
or on both sides, but there is closure at the mid line. The differences 
between these types may be illustrated by the diagrams showing 
the areas on the roof of the mouth against which the tongue has 
contact for [0], [s], and [1]. Other types and many intergradations 
of these are possible. 

14.8 There are numerous types of stops which can profitably 
be distinguished, and the mechanisms for many of them are rela¬ 
tively complex. The following descriptions of types of stops will be 
largely restricted to the opening phase (the closing phase is gen¬ 
erally comparable, but variations are less frequently phonemically 
significant), and assumes that the stop will be followed immediately 
by a vowel [a]. 

A voiced stop is one in which the cords are in position for voice 
before the opening of the stop. The voice then starts as soon as 

Articulatory Phonetics 195 

the air stream starts. This may precede the opening by a short 
interval, so that if it is not held too long, voice may actually con¬ 
tinue through the stop. A voiceless stop is one in which the cords 
are relaxed as the air stream starts, so that the start of the voice 
is delayed. The length of this delay may materially affect the per¬ 
ceived quality of the consonant. The air pressure necessary to 
produce voice for the following vowel will produce a strong puff of 
air in the interval between the opening of the stop and the closing 
of the vocal cords. The result is a voiceless aspirated stop. Since 
the puff of air (aspiration) is usually accompanied by glottal fric¬ 
tion it may be represented by [h] and the sequence by [tha] or 
[t h a], etc. 

If the build-up of air pressure is delayed until the vocal cords 
are brought into voicing position, there will be no puff of air. 
Such a stop is a voiceless unaspirated stop, or simple stop. Un¬ 
aspirated stops may be indicated, when the lack of aspiration is to 
be emphasized, by [t“], etc. Aspiration is a more or less relative 
matter. All intergradations are possible. The terms aspirated and 
unaspirated imply that only two values are considered significant. 
For some linguistic purposes much more precise specification is 

Aspiration may usually be detected by holding a thin strip of 
paper in front of the lips. If aspiration is present, the strip will 
jump noticeably outward. Of course, the strip must be adjusted 
to the right degree of flexibility. Aspiration may also sometimes 
be felt against the back of the hand. These methods depend on 
the fact that aspiration involves a puff of breath, but this is usually 
closely correlated with the acoustic impression. 

Voiced aspirated stops are much less frequent than voiceless 
aspirates. There must be voice or some similar sound at the mo¬ 
ment of opening, followed by a puff of air. This is usually acousti¬ 
cally similar to [fi]. The sequence [dim] differs from [tha] in that 
there is voice or relatively strong glottal friction throughout. 

14.9 Stops, and less often other types of sounds, may also be 
classified on the basis of the strength of articulation. Fortis sounds 
are those produced with relatively stronger articulation, lenis with 
relatively weaker. Voice, aspiration, and strength of articulation 
often combine in various ways. In English /ptk/ are generally 
fortis, voiceless, and aspirated, /b d g/ are generally, lenis voiced, 

196 Articulatory Phonetics 

and unaspirated. All three variables are relative, and various 
intergradations occur. Moreover some allophones of either set 
show different combinations. 

Stops can be pronounced with more or less simultaneous glottal 
closure. Such sounds are said to be glottalized. Glottalized stops 
are most commonly very fortis, and glottalization is sometimes 
thought of as merely the extreme of the fortis : lenis dimension. 
However, glotallizcd lenis sounds are used in certain languages. 

14.10 Affricated stops or affricates are produced by a relatively 
slower opening, simple stops by a relatively faster opening. As the 
stop is released it is necessary to pass through an articulation 
which, if held, would produce a typical fricative. For the time 
interval in which the articulators are passing through this position, 
friction is produced, and this contributes to the total acoustic im¬ 
pression which is heard as a stop. Affricates and simple stops differ 
in the prominence of this element of friction. An affricate is in 
some respects the same as a stop plus a homorganic fricative. 
Whether it is heard as a single sound or as a cluster depends on the 
phonemic patterns of the hearer. 

14.11 There are as many types of affricates as there are frica¬ 
tives. The chart shows two: median and lateral. The dental, 
alveolar, and alveopalatal median affricates are most commonly 
grooved. ft] - [ts], ft] = [tS], [J] = [d*]. Slit affricates such as 
[tO] and [pf] also occur, though not given in the chart. The com¬ 
monest lateral affricates are [X] = [tl] and [X] = [dl]. These may 
also be described as laterally released stops, since the distinction 
between [t] and [X] rests as much on the fact that the latter is 
released first at the side rather than at the center as on any differ¬ 
ence of speed. 

14.12 Since an affricate differs from a cluster only in phonemic 
status, it might be thought that to discuss affricates in a treatment 
of phonetics is a confusion of phonemic and phonetic facts. In a 
sense this is so, but no more so in this instance than in any treat¬ 
ment of sound phonetically as a sequence of segments. The seg¬ 
mentation of speech is strongly affected by phonemic patterns. 
Phonetically there is no basis for the kind of segmentation which 
we customarily use. 

Moreover, human hearing seems to be such that our impression 
of speech as a sequence of segments cannot be other than pho- 

Articulatory Phonetics 197 

nemic. In even the slowest affricate the fricative element is too 
short to be heard separately. The whole sequence is heard as one 
acoustic effect; it is interpreted on the basis of phonemic patterns 
either as one or as two sounds. This is not peculiar to affricates; 
it is just as true of aspirated stops. It is also true of most consonant 
clusters, or even of sequences of consonant and vowel. l"he shortest 
stretch of sound which can be heard as separate is longer than most 
of the segments which are heard as distinct sounds in normal 
speech. What is heard is a sequence of sounds each more or less 
smeared over its neighbors. The division of the stream of sound 
into discrete parts is a process of interpretation. There is no funda¬ 
mental difference between the instances, like affricates, where the 
segmentation differs conspicuously from language to language, and 
other sound features which are generally interpreted much the 
same in many languages. 

14.13 Variation in the manner of release of stops can be 
matched by variation in the manner of closure. Some languages 
have preaspirated stops. Perhaps more common are prenasalized 
stops. In these the oral closure slightly precedes the velic closure. 
The result impresses Americans often as a stop preceded by a short 
homorganic nasal [ m b], [ n t], etc. As with different types of release 
the interpretation of such sounds as single sounds or clusters de¬ 
pends on phonemic patterns of the language. 

14.14 Implosive stops are made by drawing air into the phar¬ 
ynx by closing the cords and pulling the larynx downward, thus 
producing a slight vacuum above it. When the stop is released 
there is a very slight movement of air inward. This may be demon¬ 
strable by holding a paper strip before the lips; often, however, it 
is extremely weak. Voiced implosives result from an incomplete 
closure of the cords. In this case there may be no movement of air 
inward on release, because the vacuum in the pharynx and mouth 
has been expended in voicing. 

14.16 Resonants are sounds in which the only function of the 
mouth and nose is to modify by resonance the sound which is 
already produced in the larynx. That is, there is no constriction 
in the mouth narrow enough to produce friction. There must be 
an unobstructed passage outward from the larynx. This may be 
through the mouth, through the nose, or through both. If only 
the nose is open, the sound is a nasal. The mouth is then a dead- 

198 Articulatory Phonetics 

end cavity, but its size and shape effect the resonance. Nasals are, 
therefore, best classified by the position of the closure in the 
mouth, a nasal being possible for any articulations for which a 
stop is possible, provided only that the closure is far enough for¬ 
ward to leave the velic passage open. 

If both the nasal and oral passages are open, the sound is 
nasalized. This type is not provided for in the chart, since the 
sounds are conveniently considered as modifications of the cor¬ 
responding oral sounds. They may be symbolized by [‘3 written 
over the symbol for the corresponding oral sound. Any oral 
resonant can easily be nasalized; oral fricatives are sometimes 
nasalized, but stops cannot be, of course, since the opening of the 
nasal passage would immediately disqualify the sound as a stop. 

14.16 If only the mouth is open, the sound is an oral resonant. 
These may be classified as median and lateral. Lateral resonants 
are closed at the mid line, and usually at one side. The other side 
is open enough that no friction occurs. Most laterals have an [1J- 
like quality to American cars, though there may be several con¬ 
trasting laterals in some languages. 

Median resonants include most vowels and various vowel-like 
consonants. The latter are commonly called semi-vowels. Whether 
median resonants are to be considered as consonants or as vowels 
is a matter of phonemic function in the particular language, not of 
phonetic nature. The three that arc shown on the consonant chart 
are the three that occur in the consonant system of English. Only 
tradition and English background justify listing these in a general 
consonant chart. English /y/ is any mid or high palatal resonant 
functioning as a consonant; /w/ is any mid or high velar resonant 
with strong or moderate lip rounding functioning as a consonant; 
/r/ is in many dialects any median resonant with tongue-tip 
retroflexion. All three are difficult to describe phonetically because 
of numerous allophoncs and because the phonemic problem of the 
consonant: vowel contrast (which is phonetically irrelevant) looms 
so large in our thinking about them. 

14.17 Before continuing with the discussion of median reso- 
nants in their aspect as vowels, it is convenient to sum up the 
description of consonants by presenting a chart showing some of 
the more common sounds classified on the two bases which have 
been presented in the last several sections. This chart is definitely 

Articulatory Phonetics 199 

not an inventory of all the consonant sounds possible, or even of 
all used in languages. Such a listing is not feasible, and probably 
would be of very little use if it were made. This is a skeleton chart 
in that most of the blank spaces represent possible sounds. (Not 
all! a glottal nasal, for example, is impossible.) It is also partial in 
the sense that many other sound types and other articulations are 

For example, the whole system of clicks is omitted. These are 
produced by closing the oral passage front and back, drawing the 
tongue downward to produce a vacuum, and releasing at some 
point. The most common are dental stop-like, alveolar affricate¬ 
like, and lateral. To include them in a phonetic chart involves more 
than merely adding a few more rows. They constitute a whole 
system of sounds coordinate with that exhibited in this chart. 
Clicks are phonemic in a few languages, but in many more are used 
in a few special words, as signals to animals, or as exclamatory 

The chart also omits most types of double articulations. In 
addition to the main articulation on the basis of which sounds are 
classified in this chart, they frequently have one of several other 
constrictions, usually less close, which modifies the sound in some 
discernible way. One of the most important is labialization (the 
sound is said to be labialized), in which there is added lip rounding. 
Another is palatalization, in which there is an added constriction 
between the tongue and palate. Customarily phoneticians con¬ 
sider these sound types as produced by modification of some other 
sound type, and they usually transcribe them by the use of some 
sort of diacritic. In the examples and problems in this book we have 
marked labialized sounds by a raised £*], c.g., [k w ] and palatalized 
by a raised [ y ], e.g., [k y ]. 

The chart does, however, include two common sound types 
which do not easily fit into the basic division into stops, fricatives, 
and resonants. These arc trills and flaps. A trill is a rapid alterna¬ 
tion of two homorganic sounds, one being more open than the 
other. It is produced by the vibration of some flexible organ (lips, 
tongue, uvula), but the vibration is too slow to have an identifiable 
pitch. A flap is produced by a very rapid motion of an articulator. 
Often it is comparable to a single vibration of a trill. Sometimes it 
is best described as an exceedingly short stop. 

200 Articulatory Phonetics 


v!. = voiceless vd. = voiced. 

This chart shows only some of the more important points of articulation 
and types of consonants. Symbols are provided for only a small part of those 
shown. Most of the blank spaces represent pronounceable consonants. 

Articulatory Phonetics 201 

14.18 Median oral resonants are frequently referred to loosely 
as vowels. Properly the term should be reserved for use in pho¬ 
nemic description, but it is convenient to use the short term in 
phonetic contexts. Different qualities of vowels are produced by 
variations of the following types. Vowels may be described or 
defined in terms of these variables: 

1) The height of the tongue 

2) The position (front or back) of the highest part of the tongue 

3) The position of the lips 

4) The openness of the nasal passage 

5) The shape of the upper surface of the tongue 

6) The tenseness of the muscles of the tongue 

7) Various phenomena in the pharynx and larynx 

The first two of these seem to be nearly universally significant, and 
hence serve as the basis for the customary primary classifications 
of the vowels. The others are treated as producing modifications of 
the basic vowels. This is, of course, in part at least a reflection 
of the phonemic structure of European languages. 

14.19 There is complete intergradation in all these character¬ 
istics by which vowels can be described and classified. As a result 
there is no possibility of complete agreement on certain vowels as 
basic. Various phoneticians and linguists have used different frame¬ 
works for the classification of vowels. The differences are generally 
more evident than significant, but can be puzzling to students. In 
any case, vowels can be precisely defined only by reference to some 
arbitrarily selected standard set of vowels. These cannot be speci¬ 
fied by written description, but only by oral transmission, either by 
personal teaching of a phonetician or by recordings, carefully made 
and equally carefully played. 

14.20 One such system is that proposed by Daniel Jones. This 
is a set of eight cardinal vowels. No. 1 is defined as the extreme 
high front unrounded vowel, No. 4 as the extreme low front un¬ 
rounded vowel, and Nos. 2 and 3 are interpolated in such a way 
that the intervals between seem to be “equal” in terms of acoustic 
impression. The four back cardinal vowels are similarly defined, 
but all are rounded. The eight cardinal vowels are frequently in¬ 
dicated on a vowel triangle (actually usually drawn as a trapezoid). 

The cardinal vowel system is frequently referred to in works by 

202 Articulatory Phonetics 

British linguists. It can be used with precision only by one spe¬ 
cially trained in it, but for others, the definition of vowels in 
terms of the cardinal vowels often serves to give a good rough idea. 
The vowel systems of languages are frequently presented in the 

i a 10 
,v It 
(0 u 
ID n 

The Cardinal Vowels 

English Vowels 

form of such vowel triangles with the positions marked. This is 
done in the figure for the allophones of American vowels most fre¬ 
quently heard as simple vowels. (The allophones before /y w h/ 
arc generally different. Note that in no case do the American 
vowels coincide with the cardinal vowels.) 

Not infrequently eight more cardinal vowels are listed. These 
have the same tongue positions as those defined above; however, 
the front vowels (9-12) are rounded and the back vowels (13-16) 
are unrounded. These are given with the symbols customarily used 
in Britain, the so-called IPA (International Phonetic Alphabet) 

14.21 Another system, commonly used in America, employs 
most of the same symbols with very nearly the same values but 
with less claim to high precision. In addition a number of other 
symbols are added. This system also provides symbols for the 
central vowels, lacking in the British system of cardinal vowels, 
though provided for elsewhere in the IPA notation. In the chart 
on page 203 only those symbols are given which are used in this 
book or the workbook or are important for some other reason. 

The lower high, mean mid, and higher low vowels are defined as 
being lax, that is pronounced with less tension on the muscles of the 
articulators. The remaining vowels are defined as tense. 

14.22 Diphthongs may be considered either as vowels in which 
there is appreciable change of quality during the course of their 
pronunciation, or as sequences of vowels or of vowels and semi¬ 
vowels. Phonetically the first interpretation is generally best; 
phonemically they are often best treated as sequences, in other 

Articulatory Phonetics 203 












§ I 





a a 




i u 

Lower high 



Higher mid 



Mean mid 



Lower mid 




Higher low te 

Low a 

instances as single phonemes. Thus there may be a marked differ¬ 
ence in the phonetic and phonemic significance of such a term as 

The stress may continue evenly through a diphthong, in which 
case it is called a level diphthong, but more often it will vary 
perceptibly. Those in which the stress is strongest at or near the 
beginning are called falling diphthongs. (The stress falls.) Those 
in which the main stress is at or near the end are rising diphthongs. 
English /ay aw oy/ are falling diphthongs with the less stressed 
clement phonemically interpreted as a semivowel, /yu/ is pho¬ 
netically a rising diphthong, though not usually treated as pho¬ 
nemically a diphthong in English. Most often, but not always, the 
stress is strongest on the lower vowel and weakest on the high or 
central vowel. However, some languages have phonemic contrasts 
between rising and falling diphthongs that are otherwise quite 

14.23 Phonetically, speech is always something more than a 
linear succession of sounds. Since these are mostly produced by 
air expelled from the lungs, the respiratory apparatus in the thorax 
necessarily breaks the sequence up into portions. The most obvious 
of these is a breath-group. This is the chain of sounds produced on 
one breath. Its maximum duration is controlled by the necessity 
of periodic inhalation. A breath-group does not, however, neces¬ 
sarily last as long as the air contained in the lungs might allow. 

There are two partially independent mechanisms which control 

204 Articulatory Phonetics 

inhalation and exhalation of air. The first of these consists of the 
diaphragm and the abdominal muscles. These vary the volume of 
the thoracic cavity by moving its lower wall (the diaphragm) up 
and down. They seem to move more or less steadily throughout 
each breath-group, normally reversing their action between breath- 
groups for inhalation. This constitutes, therefore, the physio¬ 
logical basis of the breath-groups. 

The second breathing mechanism consists of the intercostal 
muscles. These extend between successive pairs of ribs, and increase 
or decrease the volume of the thoracic cavity by moving the side 
walls (the rib-case). In speech the activity of the intercostal 
muscles does not continue steadily through the breath-group, but 
is subject to more rapid variation. This correlates in the simplest 
case with the alternation of vowels requiring relatively large 
amounts of air with consonants requiring less. Speech is, therefore, 
marked by a scries of short pulses produced by this motion of the 
intercostal muscles. These pulses are the phonetic syllables. Typ¬ 
ically a syllable centers around some vowel or other resonant and 
begins and ends in some sound with relatively closed articulation. 

All speech consists of a sequence of such syllables and breath- 
groups, which are phonetically the basic framework of speech and 
the most clearly detectable segmentation. Their phonemic status 
is, however, another matter. In many languages syllables have no 
phonemic status whatever. They arc merely part of the phonetic 
mechanism by which pronunciation is effected. In others, the rela¬ 
tionship of sounds to these syllable pulses is phonemically signifi¬ 
cant, so that syllable division must be taken into account in a 
description of the language. Even in the latter case, however, it 
does not follow that phonetic and phonemic syllables are iden¬ 
tical, and to assume that they are can be productive of serious 
error. In this as elsewhere, phonetics and phonemics must be 
kept distinct, though obviously they are intimately interrelated 
in many ways. 




Acoustic Phonetics 

15.1 The theoretic basis of music has been understood for a very 
long time. It has long been believed that somehow speech must be 
subject to much the same basic principles, but until recently no 
one has been able to show the relationship clearly. In very recent 
years new instruments have been developed which make it possible 
to record and measure various features of sound waves in such a 
way that the significant features of quality can be correlated with 
known facts about the physical basis of music. Both can now be 
brought within one comprehending physical theory. Of course, such 
a theory is no more able to explain speech than it is to explain 
music. It can merely describe the physical features of sound which 
are used in speech and music and demonstrate that they are 
basically similar. How they are used is a matter for study by the 
special methods of musicology and linguistics. 

These new developments obviously hold great theoretical im¬ 
portance to linguistics, particularly for phonology. The revolution 
in thinking which they seem to be stimulating promises to be 
profound. But they have much more than just theoretic interest, 
since they have provided many new and powerful tools for the 
analysis of certain phonetic features of great importance in 
language. Some knowledge of the basic principles of acoustic 
phonetics is becoming essential for an understanding of modern 


206 Acoustic Phonetics 

16.2 An understanding of the elementary physical principles 
behind music is essential for any understanding of the nature of 
these new developments. While many of the students who use this 
book have some acquaintance with them, it is necessary first to 
present some very elementary principles for those whose back¬ 
ground does not include either acoustics or music theory. 

The physically simplest sound is one such as that approximated 
closely by a high-grade tuning fork and known as a pure tone. A 
properly designed fork will vibrate with great accuracy at a 
constant frequency. This frequency can best be stated in cycles 
per second, abbreviated cps. One cycle is the complete movement 
from some fixed point to one side, thence back through the starting 
point to the other side, and thence back to the starting point. 
Such movement is continued with gradually diminishing strength 
until the energy imparted by striking has been expended. The 
count of the cycles completed in one second is the measure of the 
frequency. The frequency is one convenient measure of the pitch 
of the sound which is produced. Or, as a tuning fork can produce 
only one tone, it is sometimes stated as the pitch of the fork itself. 
Cycles per second is the measure of pitch preferred by physicists. 

Another convenient and familiar 

measure is in terms of the 

musical scale. The pitches used in conventional Western music are 
designated by the letters A to G, or by letters with sharps or flats 
added. The piano keyboard contains eight notes designated A. 
Under one system of tuning, these have the following frequencies: 

First note on Keyboard 

27.5 cps. 



A below Middle C 


A above Middle C 





Each of these notes differs from those above and below it by an 
octave. In frequency each is exactly double that below it. A differ¬ 
ence of pitch of two octaves is equal to a multiplying of the fre¬ 
quency by 4 or three octaves to multiplication by 8 or etc. 

Acoustic Phonetics 207 

All notes designated by the same letter differ by an integral 
number of octaves. 

15.3 If A (220) and A (440) are sounded together the combina¬ 
tion is said to be harmonious. This is the acoustic impression when¬ 
ever there is a simple mathematical relationship between the 
frequencies. In this case the ratio is 1:2. If A (440) and E (059.2) 
are struck together the resulting sound is also harmonious. In this 
case the ratio closely approximates 2:3. That this is only approxi¬ 
mate is the result of compromises which are made in order to 
enable a piano to be played in more than one key. If on the other 
hand A (440) and A flat (415.3) arc played together the sound is 
unharraonious. There is no simple mathematical ratio between the 
two frequencies. 

All notes having frequencies which are exact multiples of a given 
frequency are said to be harmonics of it. The basic note in such a 
series of harmonics is called the fundamental. There are various 
ways of designating harmonics. We shall use numerals representing 
the multiplier. For example, A (440) is the 2nd harmonic of 
A (220). 

16.4 While a tuning fork produces a very nearly pure tone, 
few other instruments do. Typically a musical instrument pro¬ 
duces a fundamental frequency and a whole series of harmonics. 
The relative strength of these harmonics determines the sound 
quality of the instrument. This may be demonstrated by an 
electric organ. This instrument can produce a close approximation 
to a pure tone. It also produces a series of harmonics. The organist 
can produce various effects by mixing the fundamental and various 
harmonics in different proportions. In this way various instruments 
can be imitated and other new and strange sound effects produced. 
Criticisms of the tone quality of the electronic organ arise in part 
from the fact that only a relatively short series of harmonics is so 
used. Actually tone quality is affected in some way by every 
audible harmonic. 

This can be a very large number. Consider a bass voice pitched 
at A (110). The following frequencies will be produced: 

Note sung A 110 cps. Fundamental 

1 octave higher A 220 cps. 2nd harmonic 

E 330 cps. 3rd harmonic 

208 Acoustic Phonetics 

2 octaves higher 


440 cps. 

4th harmonic 

C sharp 

550 cps. 

5th harmonic 


660 cps. 

6th harmonic 

about G 

770 cps. 

7th harmonic 

3 octaves higher 


880 cps. 

8th harmonic 

9th, 10th, 11th, 12th, 
13th, 14th, 15th 

4 octaves higher 


1760 cps. 

16 th harmonic 
17th-31st harmonics 
(15 in all) 

5 octaves higher 


3520 cps. 

32nd harmonic 
33rd-63rd harmonics 
(31 in all) 

6 octaves higher 


7040 cps. 

64th harmonic 
65th-127th harmonics 
(63 in all) 

7 octaves higher 


14080 cps. 

128 th harmonic 

Somewhere in the next octave (20,000-25,000 cps.), the limit of 
human hearing is reached. In this example it might be at about 
the 200th harmonic. The actual limit varies from individual to 
individual. The higher audible frequencies are of little importance 
in speech. Home tape recorders are generally designed to record 
up to about 8000 cps., dictating machines perhaps 4000 cps. The 
latter record speech intelligibly, but hardly naturally, and do 
occasionally interfere in small ways with understanding. 

15.6 The second feature which must be stated to describe the 
vibration of a tuning fork is the amount of movement which there 
is in each cycle. The most familiar measure of this is the amplitude. 
We will not attempt an exact definition, as to do so would lead us 
further than necessary into physics. However, amplitude is related 
in a loose way to loudness. After a tuning fork is struck the note 
gradually fades away into silence. This drop in loudness is the 
psychologic response to a decrease in amplitude. 

If two tones are sounding at once the frequency and amplitude 
of each can be stated. The quality of the sound heard will depend 
on the two frequencies and the relative amplitude, the loudness of 
the sound on the absolute amplitudes. As linguists we are interested 
primarily in sound quality. Our only direct interest in loudness is 

Acoustic Phonetics 209 

in relative loudnesses (an approximation to stress) of successive 
sounds. We, therefore, have little interest in amplitude as such, 
but are much concerned about relative amplitude. 

The physicist would add one more variable, phase, by which he 
would specify any complex sound. Like absolute amplitude, this 
has no linguistic significance, though it does have importance in 
some non-linguistic aspects of human hearing. We mention the 
matter only for the benefit of those who have studied physics, and 
we need not attempt a definition. 

These two variables, frequency and relative amplitudes, are 
adequate to describe the quality of any steady continuing sound. 
Such sounds are not characteristic of speech, where the sound 
quality seems to be continuously varying. It is therefore necessary 
to state the changes of frequencies and relative amplitudes 
with time. The time dimension is fundamental in all linguistic 

15.6 All that is linguistically significant (and a lot more!) about 
a given sound (say an approximately steady portion selected from 
an utterance) can be stated by listing the frequencies which enter 
into its composition and the relative amplitudes of each. Such an 
analysis of sound is generally best presented graphically, in which 
case it is called a spectrogram. Sound spectrograms can be produced 
by an electronic device known as a sound spectrograph. This is the 
instrument which is basic to the new advances in acoustic pho¬ 
netics, though many other instruments have also made important 

Actually, it is physically impossible for a spectrograph to pro¬ 
duce anything more than an approximation to a true spectrogram. 
It is not possible to make an exact measurement of the amplitude 
at an exact frequency. Instead it is necessary to measure the 
amplitude of any and all frequencies which fall within a band of 
frequencies. In one setting of a familiar type of spectrograph the 
width of the band is 45 cps. Very roughly one band extends from 
0 to 45 cps., the next from 45 to 90 cps., and so forth through the 
range of frequencies of interest in speech analysis. In the dia¬ 
gram on page 210 a crude schematic spectrogram is presented. 
The sound represented is approximately [e] said at the pitch of 
A (220 cps.). The strongest harmonic is the third, and the ampli¬ 
tude at that frequency has been listed as 1.0. The amplitudes of 

210 Acoustic Phonetics 


"True" Character 
of Sound 

Scale of 
Bands used 


Narrow Band 

the other harmonics as far as the eleventh arc given in terms of 
fractions of the amplitude of the third harmonic. 

Since the lowest pitch involved is 220 cps. (the fundamental), 
there is no amplitude recorded in the first band (0-45 cps.), nor 
the second (45-90 cps.), third (90-135 cps.), nor fourth (135- 
180 cps.). The fifth band (180-225 cps.) includes the fundamental, 
and so some amplitude is recorded by a short line. Then there are 
several bands with zero amplitude until the band 405-450 cps. is 
reached. This contains the second harmonic, and so another short 
line is recorded — and so forth to the highest frequency in the 
diagram, 2500 cps. The result is one type of spectrogram. As may 
be seen it is only an approximation. (And in actual operation some 
factors not mentioned would enter to produce other small differ¬ 
ences from the “true” picture.) It is, however, a representation 
which, in spite of all its shortcoming, reveals a great deal about 
the sound being analyzed. 

15.7 The chief limitation of such a spectrogram is that it can 

Acoustic Phonetics 211 

be used only on samples of sound which are reasonably uniform. 
To study the change of sound quality we need to make such 
spectrograms of numerous samples, spaced a few centiseconds 
apart. These are, however, awkward to compare. Some of these 
limitations can be removed by another setting of the machine. In 
this the amplitude in each band is shown by the darkness of color¬ 
ation. Each harmonic appears as a gray spot; the spaces between 
remain white. The strongest harmonics are nearly black, the 
weakest very faint. Such a spectrogram is said to have intensity 
representation, in contrast with the linear representation spectro¬ 
gram just described. The diagram gives a schematic example of 
such a spectrogram of the same sound sample. 

The limitation of an intensity representation is clearly seen in 
the diagram. It is not possible to read off the differences in color 
with any exactness. In fact, if the machine is adjusted one way, 
all that are weak are missed entirely, though the strong and moder¬ 
ate ones are clearly differentiated. It can be adjusted to bring out 
the weaker bands, but then the moderate ones are as black as the 
machine will register, and hence not distinguishable from the 
strongest. This lack of precision and versatility in registering am¬ 
plitude is the chief drawback of this type of spectrogram. 

The advantage of intensity representation is that the whole 
spectrogram of a single short sample of sound is narrow. It is there¬ 
fore possible, in effect, to record a number of such spectra, repre¬ 
senting successive sound fractions, in a series. This introduces the 
time clement and makes it possible for the spectrogram to record 
not only the sound quality of a sample, but also the changes which 
take place in speech. 

In practice the spectrograph does not make analyses of suc¬ 
cessive portions independently. Instead the amplitude within any 
one of the bands is measured through one playing of a recording. 
A stylus marks the result as it moves across the paper. The stylus 
then moves up slightly and records again as the amplitude is 
measured in the next band. Thus a continuous spectrogram of a 
sample up to 2.4 seconds in length is obtained. The usual type of 
spectrogram used in phonetic investigations is of this type. Fre¬ 
quency is represented, as in the schematic diagram on page 210, by 
the vertical dimension on the paper. Time is represented by the 
horizontal dimension, and amplitude at any given frequency at 

212 Acoustic Phonetics 

any given time by the blackness of the image at that point. This 
complicated representation can be made by a spectrograph in only 
a few minutes after the sample is recorded magnetically on its 
drum. The operation is largely automatic. 

16.8 A variety of facts can be read off such a spectrogram. In 
the first place, the frequency of each harmonic is easily discernible. 
On our schematic spectrogram eleven harmonics are visible. (An 
actual spectrogram would usually extend to a higher frequency 
and hence show several more.) The lowest of these is in the band 
180-225 cps. (On an actual spectrogram the bands are not so 
sharply demarcated, but the principle is the same.) We therefore 
know that the fundamental frequency is somewhere in this range. 
This is not very precise, somewhere between a bit above F and a 
quarter tone above A. Such a measurement would be utterly inade¬ 
quate for musical work! But if we examine the tenth harmonic we 
find that it falls in a band 2160-2205 cps. The frequency of the 
fundamental must be one tenth of this, that is it must lie between 
216 and 220.5. This is close enough for speech work, and perhaps 
of some value in music. Measurement of a higher harmonic would 
allow yet greater precision. 

But measuring the pitch at a single point is by no means all. 
Even without measuring, the risings and fallings of the higher 
harmonics give a graphic picture of the intonation or other pitch 
phenomena of speech. Not only the phonetic expression of the 
pitch phonemes, but also some aspects of the terminals can often 
be clearly seen, and if necessary measured. 

16.9 When a spectrogram such as that in the schematic dia¬ 
gram, or an actual product of the spectrograph, is examined, it will 
be found that at most points on the time scale certain harmonics 
or groups of adjacent harmonics are strong, while others are weak. 
If the pitch is rising or falling rapidly, it will be seen that this 
strength is not a property of specific harmonics as much as of 
regions in the spectrum. If a given vowel is said with a rising in¬ 
flection, the spectrogram will show a number of regions in which 
the harmonics are strong. Each harmonic will rise with the rise in 
pitch. As it enters one of these regions it will become stronger. As it 
passes out above it will become weaker again. Such a region in a 
spectrum is called a formant. There may be only one harmonic 
within a formant, or there may be a number. Harmonics can enter 

Acoustic Phonetics 213 

or leave a formant either through pitch change, or through changes 
in the position of the formant. Diagrammatic representations of 
these possibilities are given below. 

Rising Pitch Falling Pitch 

Rising Formant Falling Formant 

16.10 The positions of the formants, particularly the first and 
second, are correlated with the qualities of vowels. Any two points 
in a spectrogram in which the formants are in the same positions 
will be found to have the same vowel quality. Any two points in 
which the formants occupy different positions will be found to 
have phonetically different vowel qualities. The gross character of 
vowels can be specified by stating the positions of the first two 
formants. This is usually done by measuring the centers of highest 
intensity. Thus a formant may be stated to be located at 600 cps. 
There may or may not be a harmonic just at this point. If the 
speaker’s voice is pitched at 120 cps. (roughly B flat), there will be. 
If it is at 110 cps. (A), there will not be. However, a formant has 
appreciable width. Harmonics within a short distance on either 
side of this center frequency will be strengthened. The formant will 
therefore be evident no matter what the fundamental pitch of the 

16.11 In order to make the formants more clearly visible, the 
sound spectrograph can be set with a band width of 300 cps. in¬ 
stead of 45 cps. The spacing of the bands is not altered, so that 
there is a considerable amount of overlapping of the bands. The 
result is that the amplitude is averaged out. It is obvious that this 
will have the effect of making individual harmonics unrecordable 
unless the fundamental pitch is above 300 cps., and not clearly 
visible even then. However, the formants will show distinctly as 
prominent dark bands on the spectrograms. Such broad band 
spectrograms are much easier to read directly, but less suitable for 
exact measurement. An indication of the relationship of a broad¬ 
band spectrogram to a narrow-band spectrogram, and to a linear 
representation, is given in the schematic diagram on page 210. 


Broad Band Spectrograms of the English Vowels Cut Out of the Context ./d-d/ 

Acoustic Phonetics 215 

Broad-band spectrograms are more commonly seen in the pub¬ 
lished literature than are narrow-band spectrograms. 

16.12 The positions of the first two formants seem to be closely 
correlated with some of the more significant phonetic features of 
vowel quality. This can be seen by examining labeled spectro¬ 
grams such as those on page 214. The following table gives approxi¬ 
mate positions characteristic of English vowels said in isolation. 
Bear in mind that these figures do not represent anything of the 
range of allophonic variation. 

[i] 400, 2100 [T] 300, 1500 [u] 450, 1000 

[e] 500, 1800 [a] 600, 1300 [o] 550, 900 

M 650, 1700 [a] 700, 1100 [o] G50, 800 

You will notice that, in general, the position of the first formant is 
correlated with the height of the vowel, and that of the second 
formant with the frontness of the vowel. 

16.13 In order to explain the significance of formants, recourse 
must be had to another basic concept of acoustics. This is reso¬ 
nance. If two tuning forks of the same frequency are placed near 
together and one is struck, the second will start vibrating. The 
sound waves from the first have excited the second. A similar effect 
can be had by striking a tuning fork near an empty vessel of the 
proper size. The air column (physicists refer to any mass of air 
which vibrates as a unit as an air column) within the vessel will 
vibrate, reinforcing the sound of the fork. This is resonance. If the 
size of the air column is changed, the frequency at which it will 
resonate will be changed. This is easily demonstrated by adding or 
removing water. Changes of shape can also effect the frequency at 
which an air column will resonate, though this is harder to demon¬ 
strate simply. 

Another characteristic of resonating air columns can be demon¬ 
strated by an experiment based on a tuning fork and a resonator 
tuned to the same frequency. These can be set up so that strong 
resonance is produced. If the exciting frequency (that produced by 
the tuning fork) or the resonating frequency of the air column is 
changed slightly so that they are no longer exactly in tune, resona- 
tion does not cease abruptly, but becomes somewhat weaker. As 
the difference in frequency increases, the resonance becomes 
progressively weaker until it becomes imperceptible. Any air col- 

216 Acoustic Phonetics 

umn resonates most efficiently at some single frequency, and with 
diminishing efficiency at adjacent frequencies. There is thus a band 
of resonance, the width of which (that is, how fast the resonance 
falls off with changes of exciting pitch) depends on many factors, 
including the shape of the air column and the nature of the vessel 
that encloses it. 

Now consider an experiment in which some sound source pro¬ 
duces a complex sound consisting of a fundamental and many 
harmonics. Near this is placed a vessel enclosing an air column 
which has been so adjusted as to resonate over a rather wide band. 
All the harmonics which fall within this band will be strengthened. 
Those near the center of the band will be strengthened most, those 
a little to either side somewhat less. The spectrogram of the result¬ 
ing sound would show' a formant in the band at which the air 
column resonates. 

16.14 The spectrogram of a vowel shows a series of formants. 
On the analogy of the experiment just described, we may assume 
that each of these formants represents the baud of resonance of 
some air column somewhere in the vocal tract. Since the frequency 
of each formant is correlated in some way with the position of the 
tongue, it follows that the resonating air columns must be in some 
way controlled by tongue position. In a vowel like [i] the high 
point of the tongue divides the mouth into two cavities, only partly 
set off from each other to be sure. Various experiments have sug¬ 
gested that the two halves of the mouth are associated with the 
two lower formants. The first formant is apparently produced by 
resonance in the throat and back of the mouth; the second by 
resonance in the front of the mouth. This is of course an over¬ 
simplification, since the shapes of these cavities are very complex 
and the resonance patterns must also be complex. 

That this has some measure of truth may be seen by graphing 
the vowel positions in terms of the two formant frequencies. Since 
the high vowels have a low first formant, we will set up a scale 
with the position of the first formant reading from top to bottom. 
Since back vowels generally have lower second formants, we will 
set up a scale for the positions of the second formant reading from 
right to left. Each vowel can then be indicated on the chart at the 
intersection of the proper two measures. The result is remarkably 
like the familiar arrangement of vowels based originally on tongue 

Acoustic Phonetics 217 

position. In the accompanjdng chart, the nine English vowels are 
so indicated. In some instances two or more allophones have been 
taken into account, hence the larger areas assigned to certain 

Frequency of Second Formant 
2400 1S00 1200 GOO 

16.16 The interpretation of various consonant sounds presents 
more complications, but the same principles will be found to ex¬ 
plain many of the observed phenomena. Consider a sequence like 
[kffik], said without aspiration. The stops will appear on a spectro¬ 
gram as periods of time without any noticeable amplitude at any 
frequency, that is, as blank sections representing silence. In the 
middle of the syllable the formants will be at the characteristic 
[a;] position. As [k] is said, the tongue is in contact with the roof 
of the mouth. To reach the [®] position, it falls rapidly, passing 
through the positions of various vowels. We may therefore expect 
that between the first [kj and the [af] the formants will curve 
rapidly through positions characteristic of certain high vowels. 
Just before the second period of silence representing the second 
[k], the formants should again shift rapidly as the tongue moves 
toward a velar closure. The observable effect will be characteristic 
of all velar sounds. The actual stop portion (silence) will be iden¬ 
tical for [p t k]. These three sounds are distinguishable on the 

218 Acoustic Phonetics 

spectrograms, and presumably in hearing, only by their effect 
on the formants of adjacent vowels, or other features of the transi¬ 
tion from consonant to vowel or vowel to consonant. We cannot 
here describe the effects, but they can be observed on any good 
spectrogram and are described in detail in the literature. 

16.16 What has just been said of stops is also true of other con¬ 
sonants. The acoustic features observable in the brief moment 
when the consonant is actually articulated are seldom adequate to 
unambiguously identify the consonant. They can be read with 
surety only by using the effects which the consonant has on the 
following or preceding vowels. From a phonetic point of view it 
is not wholly satisfactory to consider phonemes as following one 
another in a neat sequence, one being finished before the next be¬ 
gins. Instead, the end of one overlaps the beginning of the next. 
Each moment in a sample of speech may be considered as a group 
of cues, each one of which serves in some way to identify some 
sound. Some of the cues at one moment may belong to a sound 
which has already been clearly marked by cues in preceding mo¬ 
ments. That is, they are the last members of the sequence of cues 
that identifies a sound. Others may be the first members of a 
sequence that will identify a succeeding sound. That this is so 
may be demonstrated without a spectrograph by cutting a tape 
recording in two and playing each piece separately. If it is cut in 
the right place, it will be found that there are two or three sounds 
that can be clearly heard on either part of the tape. The cues which 
together make up one of these sounds have been divided between 
the two pieces, leaving enough of each in each tape to permit iden¬ 
tification, though obviously the total effect is necessarily very 

16.17 Turning aside for a moment to phonemic theory, the 
question may be asked, if sounds overlap in this way, why do we 
hear them as in a lineal sequence, as we do in the case of con¬ 
sonants and vowels? It is not enough that the centers of the series 
of cues follow one another — maybe they do, but we cannot be 
sure. Our own psycho-physical limitations would prevent our 
hearing them individually anyway. Even if the sounds did not 
overlap, the ear would smear them so that in our perception they 
would seem to overlap. • 

The answer is simply that we hear the whole complex of features 

Acoustic Phonetics 219 

together, but distinguish differences of order by differences of total 
impression. We learn to do this by much the same process as we 
learn to distinguish individual phonemes. We can only do it be¬ 
cause there are contrasts for different orders of phonemes. A 
minimal pair like cat /kiet/: act /ffikt/ is an example. 

That the order of phonemes is based on contrasts may be shown 
by considering stresses. It probably would never occur to the aver¬ 
age American to ask, even in the most vague form, the question, 
which comes first, the /«/ or the /'/ in such a word as calf The 
question is meaningless, not because the /«/ and /'/ are simul¬ 
taneous (we cannot know for certain that they are!); nor even 
because they obviously overlap, since the /k/ and /*e/ do also; but 
simply because there is no evidence by which the question can be 
answered. That is, we do not have contrasts of the type of /*k'jet/ : 
/•kie't/. If we did, the question would become important, and the 
answer would become accessible. In fact, the answer would prob¬ 
ably be quite obvious to the average American, just as it is obvious 
that cat is not /*kt®/. 

There has been considerable debate in recent years about the 
relationship of “segmental” phonemes (consonants and vowels) 
to “suprasegmental” phonemes (stresses and pitches). The tradi¬ 
tional distinction is that “segmental” items form a linear sequence 
of segments, in contrast to “suprasegmental” items which do not, 
but rather overlap the segments. This is not true at all, since 
phonetically both groups are alike in overlapping. The distinction 
is a phonemic one. Each group (and the “suprasegmental" must 
be divided into two groups in English) — consonants and vowels, 

!' - + /> and /I 2 3 4 \ /* —*/ — forms a sequence. There is a 

contrast between /"'/ and /''/, or between /2 3/ and /3 2/. The 
three sequences of phonemes cannot be condensed into one because 
we lack contrasts like /' »/: /® '/, /3 '/: /' 3/, or /3 *e/ : /*e 3/. 
The English phoneme /+/ is exceptional in that there are con¬ 
trasts between /t + / and / + 1 /, night-rate : pie trade, as well as 
/+ y; /' +/. /+/ is also a classic instance of overlapping of pho¬ 
nemes, since it is not ordinarily possible to identify any segment 
of the stream of speech as /+/» as it is with, say, /t/. /+/ is marked 
only by certain cues which accompany adjacent segments. 

16.18 One other feature of spectrograms must be briefly de¬ 
scribed. On a narrow-band representation, there will be some areas 

220 Acoustic Phonetics 

in which the regular striations representing harmonics will not be 
visible. Instead there is a mottled appearance. On broad-band 
spectrograms these areas will be filled with a general gray color. 
This represents sounds in which amplitude is measurable at all 
frequencies, or randomly distributed, instead of being organized 
into harmonics with appreciable amplitude and intervening fre¬ 
quencies with none. This is known to physicists as white noise. 
It is characteristic of fricatives, and is the type of sound known to 
articulatory phoneticians as friction. 

A voiceless fricative shows only such white noise. The various 
fricatives differ in the portions of the spectrum in which the white 
noise is concentrated, as well as in transitional effects on adjacent 
sounds. In [s] it is mostly at a high frequency, in [6] mostly 
rather low. Voiced fricatives show the same random patterns super¬ 
imposed on the harmonics. 

Whispered vowels show the same formants as spoken vowels, 
but they are formed by the reinforcement by resonance of fricative 
noise produced at the larynx. 

15.19 As we have briefly sketched, the various features which 
are heard by the ear can generally be identified on the spectro¬ 
grams. That some cannot, as yet, may be due only to the present 
imperfection of our machines, or more likely, to our lack of 
familiarity with the interpretation of the results. The observed 
patterns can be correlated closely with our previous notions of 
articulatory phonetics. The sound spectrograph shows how it is 
that the human ear can relate what is heard to the methods of 
producing the sounds. We are beginning to understand not only 
how sounds are produced, but what each method of production 
does to the sound, and hence are moving toward an understanding 
of how a person can reproduce what he hears. The latter is a basic 
problem of language, though just outside the domain of linguistics. 


aim::ri3ra!!TS3:n::i:i!']£:£si i irim!{ii!rri*Ti:-Tnnmnrrni'' 



Interpretations of English 

16.1 The literature dealing with English pronunciation is exten¬ 
sive and diverse. Most of it makes no claim to be scientific, so that 
it requires no discussion here. But even within the smaller body 
of works that purport to describe English phonology on a scientific 
basis, there are considerable and often perplexing differences. It 
seems almost as though no two authors can agree, and that few 
even approach agreement. As a practical matter it is important to 
understand these differences, since otherwise a large mass of im¬ 
portant literature is inaccessible. 

There is another reason for close attention to this diversity. The 
scientific competence of descriptive linguistics to handle the data 
of human speech can best be judged by comparing such divergent 
analyses. Unless the results are reproducible and the work of 
different linguists more or less reconcilable, no great validity can 
be claimed for the methods. English is a good subject for such a 
comparison, because a large number of investigators, representing 
various points of view within descriptive linguistics, have made 
and published analyses of this language. 

16.2 The problem is perhaps best presented by giving a sum¬ 
mary in advance and then discussing the details against this 


222 Interpretations of English Phonemics 

framework. The following five general heads subsume the more 
important of the differences which can be observed between pho¬ 
nemic analyses and the resultant transcriptions. 

(1) Differences in symbols. The selection of a symbol is dictated 
by tradition, expediency, the purposes of the work, and the ar¬ 
bitrary whim of the analyst. None of these factor's is significant in 
judging linguistic science, but any variation in symbols can be 
the source of practical confusion. 

(2) Differences in the material treated. English is spoken in 
a number of dialects, each with its own phonologic system. An 
analysis can be based on any one of these, and may accordingly 
differ from an analysis based on another dialect. But these systems 
are not wholly different, so that it is possible to base an analysis 
on a group of dialects, or on all the dialects taken together. Such 
an analysis must have fundamental differences from one based on 
any single dialect. 

(3) Differences in interpretation of the data. There are in 
English, as in most languages, a few points in the structure where 
the data does not point unambiguously to one single interpreta¬ 
tion. When the evidence approaches an even balance, it may be 
expected that both solutions will find supporters. Disagreements 
arising from this source are less significant in judging linguistic 
science than is the fact that the differences can themselves be 
predicted. A linguist examining the phonology of a language can 
identify the points at which divergences of interpretation can be 

(4) Differences in the concept of the phoneme. Linguistics, 
fortunately, has not had any one generally accepted orthodoxy, 
though within restricted circles minor systems have been rather 
rigidly promulgated. The result has been a free give-and-take and 
exploration of various methods and theoretical formulations. This 
has contributed both to immediate confusion and to ultimate 

(5) Differences in the state of linguistic research. Descrip¬ 
tive linguistics is a relatively young and very rapidly advancing 
science. (The years 1924 and 1933 are among the leading candidates 
for the birth date of American linguistics.) Certain features of 
English phonology have only recently been understood, and several 
still lie on the farther side of the frontier. Each later analysis has 

Interpretations of English Phonemics 223 

had important data available which the framers of earlier state¬ 
ments could not have known. 

16.3 The transcriptions of consonants used in ten books 1 and 
the present introduction are tabulated below. Those consonants 
which are omitted from the table are written essentially alike by 

This Book 
Smith & 













































d 3 























16.4 Perhaps the most minor, but by no means the least 
troublesome, of the differences shown in the table is in the matter 
of symbols. From a purely scientific point of view this problem is 
trivial. Symbols can be assigned arbitrarily. Some linguists have 
certainly exploited this liberty too freely. But on the whole, most 
linguists have attempted to be as consistent with past precedents 
as was practical. The difficulty has been that the precedents have 
not always been consistent, and the practical exigencies have 
sometimes been severe. 

British phoneticians (among whom are Jones and Ward, with 

1 Bloch, Bernard and George L. Trager. 1942. Outline of Linguistic Analysis. 

Bloomfield, Leonard. 1933. Language. 

Fries, Charles Carpenter. 1918. Teaching and Learning English as a Foreign 

Jones, Daniel. 1950. The Pronunciation of English. 3rd ed. 

Kenyon, John Samuel. 1940. American Pronunciation. 8th ed. 

Nida, Eugene Albert. 1949. Morphology. 2nd ed. 

Pike, Kenneth Lee. 1947. Phonemics. 

Smith, Henry Lee and George L. Trager. 1951. Outline of English Structure. 

Thomas, Charles Kenneth. 1947. An Introduction to the Phonetics of Amer¬ 
ican English. 

Ward, Ida C. 1945. The Phonetics of English. 4th ed. 

224 Interpretations of English Phonemics 

the Americans Kenyon and Thomas following them) have gen¬ 
erally adhered rather closely to the conventions of the Interna¬ 
tional Phonetic Alphabet, usually known as IPA. This is the result 
of a long process of development starting in 1888 with the forma¬ 
tion of L’Association Phon&ique Internationale. One of the orig¬ 
inal objectives of this organization was the creation of an alphabet 
which would have a distinctive symbol for every sound in human 
speech, and which would supplant the chaos of notations, then 
in use, by one internationally recognized standard. This was, of 
course, long before the development of the phoneme theory in 
anything approaching its modern form, and before European 
linguists had had much experience with “exotic” languages. As 
phonetic knowledge increased, the alphabet was expanded. This 
was not done haphazardly; each proposed symbol was given careful 
consideration by the membership and adopted or rejected by vote 
of the council. Even today, almost every issue of Le MaUre 
Phmttique, the journal of the Association, carries a discussion of 
some proposal or the announcement of a decision on some symbol. 

The IPA has one great merit. The principles which were set up 
to guide the development rejected the use of diacritics (marks 
added to letters to modify their values) in favor of entirely new 
letters which would be so designed as to harmonize with each 
other. In time, diacritics had to be taken up again, simply because 
the number of sounds for which symbols had to be supplied was 
overtaxing the ingenuity of the Association members. Neverthe¬ 
less, IPA has provided a basic repertoire of symbols that are 
simple and of pleasing design. (Of course, not all the IPA proposals 
are equally felicitous in either regard!) The symbols used by Jones, 
Kenyon, Thomas, and Ward all follow closely the recommenda¬ 
tions of IPA. The rules state, “Affricates are formally represented 
by groups of two consonants (ts, t$, ds, etc.), but, when necessary, 
ligatures are used (ts, t$, dj, etc.).” Jones and Kenyon elect the 
first alternative, Ward the second. Thomas writes /6 ]/ the same 
as Jones and Kenyon, but for a different reason. 

16.5 In the meantime, interest in American Indian languages 
(Amerind) was growing among American anthropologists. They 
worked out a notation for recording their materials, but were less 
concerned with discussing the principles of the notation and the 
design of symbols than with practical field work. Their material 

Interpretations of Enslish Phonemics 225 

had to be printed, and they tended to select such symbols as would 
be at hand in a reasonably well-equipped print shop. Small caps, 
italics, raised letters, punctuation marks, Greek letters, and the 
diacritics commonly stocked were all exploited. The result was 
far from esthetic in appearance, but it served its purpose. 

In time the influence of the IPA began to make itself felt among 
American linguists. The Americanist practice accommodated itself 
more and more to IPA usages. Certain of the older conventions 
have almost wholly disappeared, but American linguists have 
maintained a certain independence from the IPA. This has been 
reinforced by their primary interest in phonemics rather than 

Recently the typewriter has begun to exert an influence on the 
selection of symbols. (The books of Fries, Nida, Pike, and Smith 
and Trager are all reproduced from typed copy.) This is the pri¬ 
mary reason for the use of d - by Pike and Nida for /S/. Use of the 
typewriter has also contributed to the American preference for 
§ 2 6 j as against the IPA $ 5 t$ d$. 

One of the most troublesome of the differences between tran¬ 
scriptions is in the use of y and j for /y/. The IPA uses j, following 
the orthographic conventions of German and Scandinavian lan¬ 
guages as well as the modern spelling of classical Latin. The 
Americanists use y, following the precedent of English orthog¬ 
raphy. It is largely to minimize this confusion that the palatal 
affricate is usually written j instead of the simpler j. (IPA used y 
for a different purpose, the high front rounded vowel [if].) 

16.6 Aside from these differences of symbolization, there is 
practical unanimity in the analysis of the English consonant system 
with the following exceptions: 

The majority treat /£ ]/ as two unit phonemes. A few, including 
Bloch and Trager and Thomas, treat these as clusters. (Jones and 
Kenyon transcribe with two letters, but list t$ and d 5 with the 
other consonants, and apparently consider them as having the 
status of units.) There is no simple and conclusive evidence for 
either analysis. Why chooset: while shoes has been advanced as a 
* Throughout this discussion / / will be used to mark phonemic equivalents 
in the system of Chapters 2, 3, and 4 of this book. The symbols under dis¬ 
cussion will be written without bracketing. This docs not imply, of course, 
that they are not phonemic, but merely that we are discussing symbols and 
phonemic analyses rather than phonemes as such. 

226 Interpretations ol English Phonemics 

minimal pair, but in most pronunciations open transitions confuse 
the picture. But there does seem to be a growing body of rather 
involved reasoning, which, while not conclusive, does generally 
favor the unit-phoneme hypothesis. In any case, the difference in 
interpretation does not indicate a sharp difference between lin¬ 
guists, because each recognizes this problem as a relatively in¬ 
decisive case, whatever may be his judgment of the total weight of 
the evidence. 

Phonetically /hw/ is sometimes a cluster [hw] and sometimes a 
single sound [m]. It can be interpreted phonemically either as a 
unit or a cluster, but linguists are almost unanimous in preferring 
hw. Any reasoning which would seem to support the interpretation 
of /hw/ as m would seem to require a similar interpretation of 
/hy/ as $. That Ward is not consistent is perhaps partially ex¬ 
plained by the fact that both are rather unusual in the type of 
English with which she is concerned. 

The suggestion is occasionally met with that /£ 2 6 J/ be inter¬ 
preted as sy zy ty dy. This has never been worked out in detail, 
and will work, if at all, only under severe restrictions as to dialect 
and type of data. For many, suit : shoot is something like /syuwt/ : 
/Suwt/, and contrasts of /£/: /ty/ and /]/: /dy/ can be found 
in many dialects, e.g., June: dune /juwn/ : /dytiwn/. 

16.7 If these same systems are compared with regard to the 
syllable nuclei, it will be seen that they fall into two traditions 
which differ most obviously in their treatment of the so-called 
“long vowels.” The basis of the difference can be seen most clearly 
through a detailed examination of one instance; for this purpose 
we will select the nuclei of bit and beet. There are at least four 
features in which the common pronunciations differ: 

Tongue position: 





lower and backer 



not appreciable, if 
present at all, down¬ 
ward and backward 

nucleus of beet 
higher and fronter 

somewhat longer 
generally appreciable, 
very short, upward 
and forward 

These fall into two groups. The first and second are differences 
in vowel quality. If they are considered as controlling, the two 

Interpretations of English Phonemics 227 

nuclei should be transcribed by two unit symbols, commonly i and 
i. The third and fourth indicate a contrast between a short simple 
vowel and the same vowel plus an additional element; if these are 
controlling, the two nuclei should be written as i and iy or the 
like. In either case the other set of contrasts becomes an incidental 
phonetic feature. Following one analysis, tense vowels are followed 
by non-significant glides. Following the other, there is a tenser 
allophone of /i/ (and other vowels) before /y/. A transcription of 
the type of i and iy is an eat-your-cake-and-have-it-too proposition 
which cannot seriously be suggested as a phonemic analysis, 
though it may be useful as a phonetic transcription. 

16.8 The six transcriptions in the simple vowel tradition are 
as follows. The equivalents in the system of chapter three are in 
some cases only approximate because of variability in application 
of the symbols. 



















































































































































a 1 




























o l 


16.9 The first two transcriptions to be considered are based on 
a type of southern British speech called “Received Pronuncia¬ 
tion” or “RP.” Both are done by British phoneticians of very 
similar viewpoint, and in fact represent identical analyses. That 
given by Ward is a so-called “narrow transcription,” that is, one 
which records a relatively large amount of phonetic detail. That 
given by Jones is more nearly a “broad transcription.” (A purely 
“broad transcription” would use eao for Jones’ eao.) This 
school of phoneticians (they do not consider themselves as pho- 
nemicists or as linguists) is relatively little concerned with the 
concept of the phoneme. Jones’ book relegates the phoneme to the 

228 Interpretations of English Phonemics 

last chapter, a bare two pages, and in Ward’s book the subject 
is treated only little less incidentally. The nearest approach to the 
descriptive linguist’s concept of the phoneme is their “broad 
transcription.” This is viewed, however, not as a more adequate 
statement of patterning, but as a device by which the facts can 
be stated with a minimum of notation, though at a sacrifice of 
precision as compared with a “narrow transcription.” The view¬ 
point is thoroughly phonetic. When variations of pronunciation 
are discussed (and they are generally contrasted with RP as 
a norm), it is never certain whether they should be construed as 
phonemic or allophonic (neither term is used). The greatest differ¬ 
ence between these systems and those to be discussed below is, 
therefore, not so much in details of interpretation as in basic 
theoretical framework. 

One peculiarity of RP must be mentioned. The mid central 
vowel /a/ occurs only without stress. When stressed it is always 
replaced by /oh/ (Jones’ o:, Ward’s 3). In words like cup, dull, 
mother, the stressed vowel, written a by Jones, is a rather high 
variety of /a/. The British systems, then, accurately reflect the 
phonetic facts by distinguishing between o:, o, and a. 

16.10 Kenyon’s transcription — and Thomas’ differs only in 
minor details — is essentially an adaptation of the Jones and Ward 
system to one kind of Mid-Western American English, though 
three special symbols are added to facilitate remarks comparing 
other dialects. The theoretic basis is essentially the same, and the 
outlook is primarily phonetic. One peculiarity of the system arises 
from a too slavish following of the British precedent. Kenyon uses 
o, as do Jones and Ward, to represent unstressed mid central 
vowels. A vowel of identical quality does occur under stress in 
America, but following British precedent Kenyon uses a for 
stressed o. Thus both Jones and Kenyon write cup lc\p. However, 
the RP and American rendition of this word are quite different, 
both phonetically and phonemically. The British vowel is some¬ 
what lower than the American and is in fact usually a high variety 
of /a/, while the American is a variety, often rather low, of /a/. 
Thus Jones’ kAp represents /k&p/, Kenyon’s kAp, /k£p/. One 
result of this transfer of values is to provide the phoneme /a/ with 
two symbols. One, d, is used only when unstressed; the other, a, 
only when stressed. The two are in complementary distribution, 

Interpretations of English Phonemics 229 

but the obvious phonemic conclusion is not drawn; the orientation 
is too strongly phonetic. Kenyon’s discussion of the phoneme is 
limited to less than one page. 

Jones makes a clear distinction between /a/ a and /aH/ a:. 
The transfer just described leaves only one symbol for the two, 
and Kenyon writes a for both /a/ and /aH/. In many Mid-Western 
dialects the contrast between the two is not very clear, and little 
damage is done; both father fertfor and bother batfr are written with 
the same vowel. But in some dialects the two are quite different 
/f&H'Sor/ and /b&Sor/. This is one instance of a significant distinc¬ 
tion which the use of this type of transcription frequently con¬ 

Kenyon also treats /or/ as a unit. This is justifiable in much the 
same way as is treatment of /ey/ as a unit e. But for this he pro¬ 
vides two symbols, o and 3 , paralleling 0 and a, and with as little 

16.11 Fries and Pike follow Kenyon in all essentials. Their 
much closer contact with descriptive linguistics and phonemic 
theory, however, causes them to drop the superfluous symbols, 
and to free themselves from the typographic trivialities common 
to this line of development. 

One problem which these two transcriptions raise, and which 
Pike has discussed at length elsewhere, is the justification for 
treating /ay aw oy/ as diphthongs, and /iy cy uw ow/ as unit 
phonemes. This is primarily another question of interpretation of 
somewhat ambiguous data. With /ay/ the evidence points more 
clearly to a diphthongal interpretation than with /iy/. Pike feels 
that this justifies treating one set as diphthongs and the other as 
simple vowels. Most linguists consider that the difference is far 
from sufficient to allow a difference of treatment, but that either 
all long vowels should be treated as diphthongs (as in the next 
group of analyses), or all should be treated as unit phonemes. The 
latter treatment has been followed by some, but in most cases 
these linguists have continued to write /ay/ as ai, or something 
of the kind, as a digraph symbol for a unit phoneme, much as 
Kenyon treats t$. 

16.12 One of the early analyses in the diphthong tradition is 
that of Bloomfield (1933). His analysis was based on his own 
speech, which was described as more or less representative of the 

230 Interpretations of English Phonemics 

Chicago area. Naturally, many of the 36 possible nuclei are not 
provided for, as they could not readily be found in this dialect. 
Aside from some symbol differences (j for /y/ and e for /ffi/), the 
following points need comment. Bloomfield knew only one contrast 
of the type /V : Vh/, /a : aH/ in bomb : balm. He did not recog¬ 
nize the diphthongal nature of the second one, and so wrote these 
as a and a. This is quite understandable, since the analysis is not 
clear unless a series of such contrasts are available. Moreover, 
since he did not detect /o/ (which he probably used, though 
rarely), he used o for /a/ when stressed. Unstressed /o/ was 
written sometimes as e and sometimes as o. This left a diphthong 
of the type /oh/ or /oh/ without a contrasting simple vowel, and 
he accordingly wjpte it as a unit o. Bloomfield’s full list, together 
with his own key words, is as follows: 










bat /&/ 


sod /a/ 


balm /an/ 



aw r 







son, sun /a/ 




saw /oh/ 



16.13 The next important analysis in this tradition was that 
of Bloch and Trager (1941). The chief advance represented in this 
analysis was to recognize the existence of /H/-diphthongs, which 
were written with h. It is notable that the material upon which the 
analysis was based included an Eastern dialect in which /h/- 
diphthongs are common and the contrasts more conspicuous than 
in the Chicago dialect. Moreover, the authors had the benefit 
of the special perspective on a language enjoyed by non-native 
speakers. The discovery of the diphthongal nature of Bloomfield’s 
a and o eliminated the need for two of his symbols. The addition 
of a eliminated a very unsatisfactory feature of the earlier anal¬ 
ysis. Since /o/ was still not clearly recognized, o was used for 
Bloomfield’s a (/a/), thus freeing a to replace Bloomfield’s e (/se/). 
Note that several of these changes are only matters of symbols, 
but the addition of h and a were significant advances, while o con¬ 
tinued its peregrinations. The following nuclei are listed: 

Interpretations of English Phonemics 231 












bear yeah 


pat /»/ 






bar calm 


pot/ a/ 






bore law 











16.14 Trager and Bloch, 3 in a fuller treatment the preceding 
year, had pointed out that all of the combinations which were 
omitted from their table (oj, uj, iw, etc.) occur in other American 
dialects. This paper included a footnote as follows: 

We do not claim that the compartments of this table will accommo¬ 
date all the syllabic phonemes of all dialects of English, though we 
believe that the exceptions will be very few and in each dialect sta¬ 
tistically unimportant. Thus, J3B [Bloch} pronounces gonna {I’m not 
gonna do it) with a short vowel in the first syllable which is pho¬ 
netically very close to the vowel of German Sonne. Though it occurs 
nowhere else in his pronunciation of English, it must perhaps 
be reckoned as independent phoneme parallel to the six short 
vowels. .. . 

Through the next decade, two hints in this statement were fol¬ 
lowed up in a series of advances which were finally summarized 
in the important publication of Smith and Trager (1951). The 
system proposed by these authors needs no detailed discussion 
because it is identical with that presented in Chapter 3, except 
that they continued the use of h for what we have chosen to 
symbolize h. 

16.16 The first of these trends was to follow up leads such as 
Bloch’s /o/ in gonna. The result was the disentangling of /o/, /a/, 
and /»/, which were somewhat confused in Bloch and Trager; 
the addition of /o/; and most importantly the addition of /i/, 
thus producing a system of nine vowels. 

The addition of /i/ calls for further comment. This phoneme is 
notably absent from all systems described here except that of 
Smith and Trager. Yet it is exceedingly common. It runs as high 
as 17 percent of all nuclei in my speech, provided the sample has a 
moderately high average word length, (/i/ is commonest in un¬ 
stressed syllables.) It is probably reasonably common in every 

* Trager, George L., and Bernard Bloch, ‘‘The Syllabic Phonemes of Eng¬ 
lish," Language 17 : 223-246, 1941. (The footnote quoted is on page 243.) 

232 Interpretations of English Phonemics 

dialect. Yet it has not been recognized until recently. Two groups 
of facts, one about the language and one about linguists, explain 
this remarkable occurrence. 

Though /i/ is exceedingly common, the distinctions /i: a/ and 
/i : i/ carry a very low functional load. This means that minimal 
pairs are relatively infrequent, or more precisely that the contrast 
is seldom called upon to signal differences between utterances. A 
functional load of zero would imply that the distinction was non- 
phonemic. The functional load of these two contrasts, while cer¬ 
tainly not zero, is one of the lowest in the English phonologic 
system. Moreover, such minimal pairs as do occur mostly involve 
syllables that are commonly pronounced differently when re¬ 
stressed forms are given in isolation. Thus, most Americans will 
say /jist/ in He just came., but when asked How do you 'pronounce 
“just"? will answer /j£st/. I find that something like ten percent 
will pronounce children /Sfldrin/ in isolation, but about half will 
use the /*/ in a context. The result has been a general confusion 
of /i/ with /i/, /o/, and even /e/. 

Phoneticians have made much of the tendency, which is real 
enough, for unstressed vowels to lose the quality which they might 
be expected to have under stress and become more mid and more 
central. This tendency led to the quite unwarranted theory that 
all unstressed vowels become a — a theory that is sometimes 
modified to permit a limited number of other vowel qualities under 
weak stress. This theory is patently untrue. Any English nucleus 
whatever may occur unstressed, though admittedly some are quite 
rare (many are rare enough stressed!). The theory was salvaged 
by the complementary theory that o (generally distinguished from 
a) is exceedingly variable. This became more or less a fixed idea 
with many, and contributed to deafening them to the occurrence 
of /i/. Such a theory is possible because of the low functional load 
of all vowels in unstressed syllables. It is not, however, low enough 
to justify any such sweeping generalizations. The discovery of the 
phonemic contrast between /i/ and /o/ has been associated with a 
complete rethinking of the nature of unstressed vowels. 

16.16 The second trend in development of phonemic analysis 
of English syllable nuclei has been a broadening of the base. 
Bloomfield’s transcription was based on his interpretation of his 
own dialect alone. Bloch and Trager worked, jointly, in developing 

Interpretations of English Phonemics 233 

a transcription to cover the speech of both, though they repre¬ 
sented different dialect areas. Other workers brought into the 
stream additional data from various areas. Smith and Trager were 
able to base their treatment on an exceedingly wide range of ex¬ 
perience with American dialects. The result has been a gradual 
shift in the nature and objective of the statements. The 36 nuclei 
currently recognized do not represent the phonemic structure of 
any one existing form of English. What Smith and Trager have 
produced is an overall analysis of English dialects, collectively. 
This is a very different thing from a phonemic analysis of a single 

In a phonemic analysis we expect each element to contrast with 
each other element. In an overall pattern analysis, this happens in 
a special way: each pair can be found to contrast in some dialect, 
but may not contrast at all in some other dialect. For example, 
some dialects use /aw/ but not /aw/; others /sew/ but not /aw/; 
still others use both /aw/ and /sew/, but in complementary dis¬ 
tribution. In each of these cases no contrast can be established 
between /aw/ and /aw/. But there are dialects in which these 
two do contrast; in one dialect, for example, lost is /l&wst/ and 
loused is /lifewst/. If the overall analysis is to cover this dialect, as 
it must to describe American English as a whole, it must provide 
separate representation for /aw/ and /aw/, even if most dialects 
do not show a contrast, /ow/ and /aw/ are commonly without 
contrast. In some dialects the word road is pronounced /r6wd/, 
in others /r$wd/. Most Americans have learned to overlook the 
difference to the extent of recognizing another’s /r$wd/ as equiva¬ 
lent to their own /r6wd/ or vice versa. In my own speech I use 
both, more or less in free variation. However, in some dialects 
there is a contrast, so that the two must be distinguished in any 
discussion of the overall pattern. Moreover, if my own speech 
is to be seen in its proper place among American dialects, it is 
important that this variability between /r6wd/ and /rSwd/ (a 
matter in which many others are quite consistent) needs to be 
noted. An overall pattern type of analysis will be much more useful 
in comparing dialects than will be an analysis based only on the 
contrasts actually used in any one dialect. 

16.17 For certain practical purposes, a transcription of the 
Smith and Trager type can conveniently be modified, as was done 

234 Interpretations of English Phonemics 

for the textbooks prepared for the English-for-Foreigners Program 
of the American Council of Learned Societies. This method prom¬ 
ises to be of sufficient importance to warrant a brief description. 

Some phonemes cause no trouble, even when the printed mate¬ 
rials are used by teachers from a wide variety of dialectal back¬ 
grounds. For example, /i/ is pronounced in very nearly the same 
list of words in all dialects. A writing such as pit will therefore 
serve equally well for any of them. Down is transcribed dawn. 
Many teachers will pronounce it /dawn/, but many others will 
say /d&wn/. But since the teacher who says /dawn/ is likely also 
to say /fflw/ in every word which is transcribed aw, this will not 
mislead the student. He can learn to associate his teacher’s /tew/ 
pronunciation with the aw transcriptions in his textbook. If he 
learns to speak /aw/, he will speak an acceptable form of English, 
even if slightly different from that other students will learn from 
other teachers. Some Americans distinguish have from halve by 
using a diphthong in the second. These are written hav and hsfehv. 
This transcription will serve the needs of the teacher who makes a 
distinction, no matter what may be the exact nature of the dis¬ 
tinction he makes. Others can merely overlook the graphic differ¬ 
ence between a and ah. Again, the student will learn an acceptable 
form of English in either case, provided he follows the usage of his 

The most complex situation is with a group of words, log, watch, 
etc. whose pronunciation is quite variable from dialect to dialect. 
In some dialects these words are pronounced with the same vowel 
as cot, which is written a. In others the vowel in these words is 
the same as that in caught etc., which in these textbooks is tran¬ 
scribed oh. To write log, watch, etc., with either a or oh would cause 
trouble to a large number of possible teachers; they are therefore 
written by a special symbol o. The students of one teacher must 
learn that a and o are pronounced alike; those of another that oh 
and o are pronounced alike. Again, either type of pronunciation will 
yield an acceptable variety of American English. 

16.18 The transcription used in the textbooks of the English- 
for-Foreigners Program is therefore basically a compromise. Com¬ 
promise is necessary, only because the transcription is required to 
do something which a phonemic transcription cannot do — that 
is, give a usable indication (an approximation) of the pronuncia- 

Interpretations of English Phonemics 235 

tion of any dialect which a teacher is likely to speak. The teachers 
in the program will be drawn from a variety of American dialect 
areas. That this can be done without serious violence to the facts 
arises from two factors. One is the low functional load of many of 
the contrasts. Indeed, any foreigner who hopes to understand 
spoken American English must learn to equate /aw/ and /aw/ 
and other similar pairs. Without this ability, he will be able to 
understand only the dialect which he has learned from his teacher. 
The second factor is a careful selection of pronunciations which 
are current over the widest areas in America. If certain local 
peculiarities had not been avoided, the result would have been 
much less useful. 

16.19 There is another type of modification in the transcription 
used in some books in the series. For example, in the edition for 
use by Turks, /6 j § 2/ are written 5 c § j. The latter are the sym¬ 
bols in the regular Turkish orthography for phonemes rather 
similar to the English / 6 } § t/. This change makes the transcrip¬ 
tion much easier for the Turkish student to use. In textbooks of 
this kind transcription is a device to facilitate teaching. As such 
there is no need for indicating all the niceties of every detail of the 
phonemic system, and even less for subservience to irrelevant 
habits of writing. Nevertheless, behind all the modifications, the 
transcriptions are ultimately based on a detailed and thorough 
analysis of English structure. This is quite different from any 
haphazard rule-of-thumb "phonetic respelling.” 

16.20 One more transcription system needs to be noted. This 
is the one used by Nida (1949). It is in the diphthong tradition. 
Since it was used merely to transcribe his own pronunciation of a 
limited corpus, there was no need for representing a large number 
of nuclei. Since he has only one /H/-diphthong, /oh/ in bought, and 
does not list the corresponding simple vowel, it is more convenient 
to write this as a unit, o, than to introduce an otherwise unneeded 
symbol for the glide. The major inadequacy is the use of the sym¬ 
bol 0 to represent both /o/ and /i/. The transcription can be con¬ 
ceived as that of Smith and Trager (which, however, it antedates), 
modified to the needs of a single dialect. As such it would seem 
to be reasonably satisfactory, and may be taken as an example of a 
type of transcription which can be expected to become fairly 
common. The list of nuclei given is as follows: 

236 Interpretations of English Phonemics 

i fill iy feel 

e pen ey pain 

a pan 

a pot ay bite 

o bought oy coy 

u put 

0 but 

16.21 Space allows only passing mention of transcriptions of 
stress and pitch. The I PA tradition generally indicates primary 
stress /'/ by ' and “secondary” (here called tertiary) /'/ by , 
placed before the initial consonant of the syllable. Thus in Ken¬ 
yon’s transcription, '$u,meter appears for /Suwmfeykor/. 

The failure to recognize secondary stress, in the sense discussed 
in Chapter 4, arises from considering the word as the largest unit 
to be examined. Words in isolation rather seldom have / A /, and 
clear contrasts with /'/ are even rarer. This limitation on the 
data is another peculiarity of the British school of phoneticians 
headed by Daniel Jones, and is incorporated in his definition of the 

16.22 Most systems of transcription give no indication of pitch. 
This is in large part a historical matter, since the phonemic analysis 
of English intonation is a comparatively recent achievement. 
Fries and Pike describe four phonemic levels of pitch (which they 
number downward; see 4.14). Their system does not indicate the 
clause terminals; the analysis and classification of these terminals 
is still more recent, and appears in Smith and Tragcr only. Earlier 
attempts to record intonation were necessarily impressionistic. 
Some authors, notably Jones, used a series of dots and lines written 
above the transcription of consonants and vowels in a manner 
reminiscent of musical notation. 

The failure of early attempts to analyze intonation led to a con¬ 
viction that its structure was somehow basically different from 
that of the consonants and vowels. This is reflected in the refusal 
of some linguists to use “phoneme” in this connection, even now 
that contrastive units can be isolated. In turn this had the effect 
of stultifying research. Intonation was cast out as a portion of 
language structure not susceptible to exact analysis. As so often 
happens in science, progress was made only when the dogmas were 

aw about 
ow boat 
uw boot 

Interpretations of English Phonemics 237 

questioned and methods elaborated in one connection (consonants 
and vowels) were applied in a very different connection (intona¬ 
tion). That these developments should take time is hardly sur¬ 
prising. Descriptive linguistics is still a young science, and we do 
not yet know fully the limits of applicability of its methods. 

The phonemic analysis of English intonation may well be one 
of the major turning points in linguistics. The phonemic analysis 
of a system of consonants was first accomplished several millennia 
ago by the unknown inventors of the Semitic alphabet. The exten¬ 
sion of the achievement to a vowel system is nearly as old. The 
analysis of the intonation system is the most original extension of 
this age-old method accomplished by modern linguistic science. 
The phonemic analysis of English pitches has already proved 
fruitful in stimulating important'advances in our knowledge of 
other languages, and so has an importance far beyond its direct 
contribution to our understanding of our own language. 




Phonemic Systems 

17.1 English was described in Chapters 2, 3 and 4 as having a 
total of 46 phonemes. It is quite natural to ask whether this is 
more or less than the average number. A search through the 
literature will reveal that languages have been reported with totals 
varying from only a bit above a dozen (for certain Polynesian 
dialects) to nearly a hundred (for certain languages of the Cau¬ 
casus). Some reflection will, however, raise considerable doubt as 
to the validity of any such comparisons; indeed, in many lan¬ 
guages only the consonants and vowels have been adequately 
described, and there is reason to believe that every language has 
significant contrasts in stress, pitch, transitions, or terminals, and 
usually in several of these categories. Many language descriptions 
overlook these completely, even in what purports to be exhaustive 
listing of the phonemes, often as the result of earlier and inadequate 
understanding of phonologic principles and methods. The reported 
minimums are therefore certainly too low. 

Moreover, the number of phonemes found depends in part on 
the methods pursued in the analysis. Not infrequently there are 
situations in which two or more interpretations are possible. These 
usually involve problems of segmentation where there is as yet no 
universally agreed-upon rigorous method. As a result, linguists 
may disagree as to the number of phonemes in a given language — 
in some cases with about equally good justification for either con- 


Phonemic Systems 239 

elusion. In the languages which have been reported to have very 
high numbers of phonemes, there are always some phonemes which 
might alternatively be interpreted as sequences of phonemes, thus 
reducing the total appreciably. Unless identical procedures have 
been used in the analysis, there is little validity in comparing 
counts of phonemes. 

17.2 If comparisons of total numbers cannot be made, there 
may still be the possibility of comparing languages on the basis of 
the presence or absence of certain phonemes or types of phonemes. 
This has commonly been done. It has long been customary to dis¬ 
tinguish “tone languages,” meaning those in which pitch is 
significant, in contrast with those languages, like English, in which 
pitch was assumed to be insignificant. Closer investigation of 
English has completely demolished the usefulness of this distinc¬ 
tion. Indeed, at the present time there is no natural language in 
which phonemic distinctions of pitch have been proved to be 
absent. It would be extremely rash to say that pitch is significant 
in all languages, because careful examination of this facet of 
phonology has been made for only a few languages. As we shall see 
below, the real basis for the distinction of “tone languages” from 
others is a structurally significant pattern in the use of pitch 

17.3 There is another objection to comparing languages on 
the basis of the presence or absence of certain phonemes. What 
precisely is meant if one says that English, Loma, Luganda, and 
Kiowa are alike in having a /b/ phoneme? Very little, unless one 
can maintain that the /b/ of the four languages is in some respects 
the same thing. But, as we have seen, phonemes can be defined 
only in reference to a given speech form. Each of these languages 
has its own set of phonemes and of contrasts between phonemes. 
It happens that, for certain reasons, partly non-linguistic, the 
symbol /b/ has been selected to represent one member in each 
system. This fortuitous circumstance is, in the case of these four 
languages, the only link, and the comparison just quoted is 
linguistically meaningless. The English /b/ is a voiced labial stop, 
the only such phoneme in the language. The Loma /b/ is one of 
four voiced labial stops in its system, each contrasting with the 
others in some additional feature. The Luganda /b/ contains a 
voiced labial stop allophone, but also a voiced fricative allophone, 

240 Phonemic Systems 

and the latter is approximately as common as the former. The 
Kiowa /b/ is used to represent a voiced labial fricative, there being 
no voiced stop to require the use of this symbol. Our statement is 
comparable to saying, “This hat, this dress, and this pair of shoes 
are all the same size, since they are all sevens.” 

17.4 What we have just said is another expression of the non¬ 
congruence of phonemic systems which inevitably plagues us in 
any work involving two languages. It is necessarily involved in any 
possible kind of comparison. Nevertheless, there is still some value 
in comparing phonologic systems, provided care is taken to 
minimize the tendency to superficiality which common symbols 
and common terminology produce. The results of comparison ma}- 
have little or no scientific validity but great practical value in 
giving a background which will contribute to an understanding of 
language. One great advantage the experienced linguist has is that 
he has some feeling for the range of possibilities in languages. 
Though even he may occasionally meet with a system far differ¬ 
ent from anything he might have expected, his feeling for the 
possibilities is helpful in guiding his investigations and suggesting 
interpretations. Statements of comparisons are, in some measure, 
a short cut to this background. 

17.5 Our criticism of a type of crude comparison in 17.3 sug¬ 
gests a somewhat more fruitful approach: the examination of the 
types of contrasts which exist within a language and which set 
its phonemes apart. For simplicity, let us restrict our attention for 
the present to phonemes which are either stops or include stops as 
important allophoncs, or pattern in some way as stops. 

English has two series of stops, traditionally labeled “voiced” 
and “voiceless.” These labels are largely conventional, as the 
actual contrasts are much more complex. The two series differ in 
one or more of three contrasts. The first series is typically voiced, 
unaspirated, and lenis; the second voiceless, aspirated, and fortis. 
However, the voice is often quite weak in /b d g/, and there arc 
more or less voiced allophones of /p t k/. Unaspirated allophoncs 
of the voiceless series occur, and in many positions the degree of 
aspiration is quite variable. The fortis: lenis distinction may be very 
tenuous. Though the contrast is quite complex, it is well main¬ 
tained in most (not all!) of the positions in which stops can occur. 

Phonemic systems with two series of stops are quite common. 

Phonemic Systems 241 

The contrast between the two may, however, be quite different. 
For example, in French the contrast is very largely one of voice. 
Both series are unaspirate. Both are, by English standards at least, 
quite fortis. The French voiced stops are generally more strongly 
voiced than are the English, and typically voiced throughout, in 
marked contrast to the gradual onset of voicing in English initial 
stops and the gradual decrease in voicing in English final stops. 
The use of the same symbols /p t k b d g/ and the labels “ voiced” 
and “voiceless” tends to obscure the basic differences between the 
two systems. 

Another possibility is that the contrast may rest entirely or 
largely in a difference of aspiration. This seems to be the case in 
many Chinese dialects. Such systems show some resemblances to 
English, where aspiration is one of several distinguishing features 
of stops, but none at all to French, in which aspiration is not 

17.6 There is, of course, no necessity that there be two series 
of stops. In Kutenai (Idaho and British Columbia) there is only 
one series /p t k q/. These are normally voiceless, but may rarely 
be slightly voiced. Ojibwa (Great Lakes Region) has only one 
series of stops. These have both voiced and voiceless allophoncs. 
They are usually transcribed /p t c k/, using the traditionally 
voiceless symbols, but this selection is arbitrary. As the usual 
English spelling of the name indicates, the stops may be heard as 
more nearly comparable to English /b d ) g/. 

17.7 Three series of stops are not uncommon. One common 
type, represented for example by Sotho (South Africa) has a voiced 
series, a voiceless aspirate series, and a voiceless unaspirate series. 
In Kiowa (North America) the three series are all voiceless, and 
distinguished as simple, aspirate, and glottalized. Korean has been 
analyzed as having three series. One is simple, with both voiced and 
voiceless allophones. One is aspirate and voiceless. Linguists have 
not been able to reach any agreement on the description of the third. 
It has been described as very fortis, as glottalized, or as merely 
doubled. If merely doubled, it can conveniently be represented 
by a cluster of identical stops, e.g., /pp/, in which case there are 
only two scries. Or the third series can be interpreted as clusters of 
a phoneme arbitrarily symbolized by /q/ and a stop, e.g., /pq/. 

Hindi and many other languages of India are generally said to 

242 Phonemic Systems 

have four series of stops: voiceless unaspirate, voiceless aspirate, 
voiced unaspirate, and voiced aspirate. Of course, it is obviously 
possible to reduce these to two series, each of which can occur in 
clusters with a following /h/. However, most linguists have been 
in agreement with the native tradition that all four are to be 
treated as series of unit phonemes. 

Of course, if the number of phonemes is reduced by treating 
some as clusters, the differences which exist between languages are 
not removed. They are merely transformed from differences in the 
inventory of phonemes into differences in the inventory of clusters. 
There is little practical difference between saying that Hindi con¬ 
trasts with English in having four rather than two series of stops 
and saying that Hindi differs from English in having clusters of 
stops plus /h/. Different phonemic analyses are merely different 
ways of describing the structure which actually exists in the 

Finally, Sindhi (Pakistan) has been reported as having five 
series. Four are comparable to those just listed for Hindi. In 
addition there is a series of voiced glottalized or voiced implosive 
stops. These may be represented as /p p h b b h 6/, etc. 

17.8 We may find similar divergencies if we examine the con¬ 
trastive points of articulation which occur in various languages. 
We shall continue to restrict our attention to those in which stops 
occur. Probably the most frequent situation is one rather like 
English, in which there are three groups of stops, one labial, one 
apical, and one dorsal/p b td k g/. French has a system which 
is the same in broad outline, and for which the identical symbols 
are traditionally used. There are, however, very significant differ¬ 
ences in detail. In articulation, the French apicals are typically 
dental; in English they are typically alveolar, with rare dental 
allophones. Superficial similarities must not be allowed to mislead 
and obscure differences of detail. 

Two contrastive apical articulations are not infrequent. In most 
of the languages of India, for example, there are both dental and 
retroflex stops. These are near the extremes of the apical range, 
and hence are about as clearly distinguishable as any two apical 
articulations can be. In Dinka (Sudan) the contrast between dental 
and alveolar has been reported. 

Two contrastive dorsal articulations occur in many languages. 

Phonemic Systems 243 

The Kutenai example mentioned above includes such a contrast. 
Mam (Guatemala) also uses such a contrast, which, together with 
a contrast between a globalized and non-glottalized series of stops, 
produces a system quite different from English. 

Rather more rarely are there additional contrasts in the labial 
region. The commonest of these is a contrast between bilabials and 
labiovelars. The latter have double articulation, simultaneous in 
both labial and dorsal regions. This contrast is found in many 
languages of Africa, and is often reinforced by some other differ¬ 
ence, commonly more fortis articulation in the labiovelars. An 
extreme case is found in Loma (Liberia) where there is a contrast 
between three labial articulations, bilabial with /p b 6/, labiovelar 
with /kp gb/, and labiodental with /v/ (the latter is a stop, and 
contrasts with a voiced labiodental fricative). 

Conversely, one or the other of the more common articulations 
may be lacking. This is the case in Tlingit (Alaska), which has no 
labials except /w/, and this in spite of a rather large number of con¬ 
sonant phonemes. Some languages have limitations of a different 
sort. Yoruba (Nigeria) has four groups of stops in two series 
/b kp gb t d kg/. The pattern is unsymmetrical in lacking a 
labial in the voiceless series. Many African languages have a partial 
series of implosive stops. In Zulu, for example, the following stops 
are found/p’ p h b 6 t’ t h d k’ k h g/. The labial implosive occupies 
a non-symmetrical position in the system in contrast with the 
glottaiized, aspirated, and voiced series which extend through each 
of the three articulations. 

17.9 Similar observations can be carried through other portions 
of phonemic systems, and the greatest possible divergencies can be 
found between languages. There is no need to detail them. It will 
suffice to remark that every possible phonetic contrast must be 
considered as at least potentially the base for a phonemic contrast 
in some language. In some instances, the distinctions seem to an 
American to be most minute and impossible of accurate discrimina¬ 
tion; but in making such a judgment we merely reveal our own 
bias, arising out of our English background. We should remember 
that some English contrasts seem hopelessly difficult to foreigners, 
notably that between /t/ and /d/ in common pronunciations of 
latter and ladder, or that between /n/ and /nt/ in a common pro¬ 
nunciation of can and can't. 

244 Phonemic Systems 

17.10 The discussion of systems of stops in the last several 
paragraphs was couched in terms of articulatory phonetics. Our 
interest at the moment is, however, rather in phonemics than 
phonetics. Our chief excuse for using phonetic terminology is that 
it is the convenient and traditional apparatus for labeling sounds 
and classes of sounds. Still, there is a deeper justification in that 
frequently the phonemic interrelationships parallel the phonetic, 
as may be seen in English. There is no need of resorting to articula¬ 
tions to identify or define the various categories of English pho¬ 
nemes. For example, /p t k f 6 s § 6/ can be set off from all the 
other phonemes of English by the fact that they condition the 
allomorph /-s/ of {-Zij, or the allomorph /—t/ of {-Dj}, or both. 
Having been so defined, they may conveniently be labeled as 
“voiceless” consonants. Similarly, /szSifiJ/ can be identified 
as those sounds which condition the allomorph /—iz/ of 

The group defined by this structural pattern is identical with the 
phonetic group of groove fricatives plus affricates. Numerous other 
pecularities of distribution or morphophonemic alternations can 
be found to define each class of English phonemes. The terminology 
of articulatory phonetics serves as a convenient source of labels for 
many of the classes so found. 

17.11 Similar types of relationships will be found to operate in 
other languages. The phonemic structures which they reveal sel¬ 
dom flatly contradict phonetic facts, but frequently do impose 
definite interpretations on the phonetic facts. In English, there are 
numerous differences in distribution between stops and affricates. 
This is not true for Hindi. The two phonetic sound types pattern 
alike in many respects. It is, therefore, useful to consider affricates 
not as a class apart, but only as a phonetically different type of 
stop. This is phonetically justifiable. The result is a symmetrical 
pattern of twenty stop phonemes: 




















17.12 A very clear instance of distribution paralleling pho¬ 
netic classification may be seen in Kutenai. The phonemes 
/ptkqsixji/ can occur as first members of initial clusters; 

Phonemic Systems 245 

/’hlmnwy/ can occur as the second member of a CC- cluster, 
as the second or third member of CCC- clusters, or as the third 
or third and fourth members of CCCC- clusters. There are other 
comparable restrictions in clusters occurring elsewhere, /p t k q s 
1 x *£/ are all voiceless; /’hmn w y/ all either are voiced or involve 
some other activity of the vocal cords. The distribution in clusters, 
therefore, parallels a fundamental phonetic distinction. 

17.13 As the last example suggests, one interesting facet of the 
phonologic structure of a language is the patterns of sequences of 
phonemes. The consonant clusters of English constitute an excel¬ 
lent example. It is relatively simple to make a complete list of those 
found at the beginning of utterances in a formal pronunciation, 
but in very informal pronunciations the problem becomes very 
much more complex. Excluding those found only in proper names, 
I have a total of 34 of the form CC- and 8 of the form CCC-. Other 
Americans (for example, those who say /tyuwn/ tune) have a 
slightly different list. A bare enumeration does not reveal much, 
however. More significant is an analysis of the formation of these 
clusters. One possible way to make such an analysis is to classify 
the consonant phonemes of English on the basis of their occurrence 
in utterance initial: 

Never in word initial /q 2/ 

Only alone in word initial /v $ z 6 J/ 

Either alone or in clusters in word initial 
Only as first member of the cluster /b d g s s h 6/ 

Only as the last member of the cluster /y wrlmn/ 
As first, middle, or last member /p t k if 

Such a classification might be carried much further, and will 
reveal that there is hardly a pair of English phonemes which can 
be used in identical lists of contexts. For example, /pr- tr- kr- pl- 
kl—/ occur, but not /*tl-/. This peculiarity is shared by ft/ 
and /d/ : /br- dr- kr- bl- gl—/ occur but not /*dl—/. Phonetic 

similarities between /t/ and /d/ are paralleled by phonologic char¬ 
acteristics. It is notable how often, even in the brief tabulation 
above, similar parallels can be found. 

17.14 The clustering patterns of other languages may be quite 
different from that of English. At one extreme is a number of lan¬ 
guages, including most of the Polynesian group, in which no 

246 Phonemic Systems 

consonant clusters occur. At the other extreme are some languages 
which use sequences of consonants which seem utterly impossible 
to speakers of English. In Cceur d’Alene (Idaho) 81 CC- initial 
clusters were reported, 41 CCO, and 2 CCCO. Final clusters are 
even more numerous: 192 -CC, 74 -CCC, 13 -CCCC, and 
/-*’?stxV were found. In Bella Coola (British Columbia) numer¬ 
ous whole words without vowels are reported: /tmk’mlp/ ‘jack 
pine tree/ /sk’lxlxc/ ‘I’m getting cold/ /ik"’t?V ‘make it big!’ 
These indicate a fundamentally different phonologic structure 
from the one familiar to English speakers. 

17.16 Not only do languages differ in the number and size of 
clusters, but they also differ in the details. For example, the pho¬ 
nemic system of English and that of Serbo-Croatian have some 
sounds which can be very broadly interpreted as the same. Each 
also has certain sounds which cannot be identified at all with 
phonemes in the other. In comparing the two it seems desirable to 
give the English /y/ and the Serbian /y/ special treatment, as 
there are some complications in each instance. The results of a 
comparison of initial clusters is as follows: 

Common to both: pr- pi- sp- br- bl- tr- st- 
dr- kr- kl- sk- gr- gl- fr- sf- si- sm- 5m- 
sn— spr- spl- str- skr- skl- 
English only: fl- §r- 

Serbo-Croatian only: pt- tk- bd- gd- Sp- 5t- 
5k- 2b- 2d- 2g- zb- zd- zg- tm- km- pn- 
kn- dm- gm- dn- gn- mr- ml- mn- ps- 
p5- p6- tv- kv- gv- 6v- £1- 6m- }b- sv- 
sh- sr- §6- §v- 51- tl- dl- zv- zr- hm- zl- 
2v- 21- 2m- 2n- vr- vl- zm- zn- ht- hv- 
hr- hi- svr- smr- Str- Skr- zdr- zgr- 2dr- 
stv- skv- zdv- svl- 2gl- 
/y/ clusters excluded 

Serbo-Croatian clusters containing pho¬ 
nemes without English counterparts 
English clusters containing phonemes with¬ 
out Serbo-Croatian counterparts 















Phonemic Systems 247 

There is, of course, no reason why English should have /fl-/ 
but not /vl-/, any more than that Serbo-Croatian should have 
/vl-/ and not /fl-/. These matters are just features of the struc¬ 
ture of the two languages. 

If a similar examination of the final clusters is made, the results 
are very different. Only /-st -§t -zd -id/ are common in Serbo- 
Croatian. In addition /-nt-nd -rjf -ys -ps -lm/ occur in some 
partially assimilated loan words. This is in very sharp contrast 
to English, where final clusters are both numerous and various. 

17.16 So far we have been discussing the structural relation¬ 
ships within consonantal systems. The same thing can be done 
for vowels, in English and in most of the other languages of the 
world. Some hints of the possibilities were given in 12.14. The 
interrelationship of the consonants and vowels, however, presents 
a different set of problems. The presentations in Chapters 2 
and 3 of these two portions of the English phonemic system were 
almost completely independent of each other. No attempt was 
made there, nor since, to prove that there is a phonemic contrast 
between any individual consonant and any individual vowel. It is 
very doubtful that this can be done in English, and certainly it 
cannot be done by any method such as that used in Chapters 2 
and 3. The nearest we can come to minimal pairs is contrasting 
words such as stone /st6wn/ : alone /5t6wn/. These are, however, 
not strictly minimal pairs, because the /o/ is associated here 
and in every similar word with /’/ or some other stress, whereas 
/s/ never is. This means that /s/ and /o/ are in complementary 
distribution. There are two reasons for not classing them as 
allophones of a single phoneme: first, there is no phonetic sim¬ 
ilarity; second, almost any consonant can be shown to be in com¬ 
plementary distribution with any vowel. 

Consonants and vowels do not contrast individually because 
they occupy fundamentally different places in the total structure 
of English utterances. This basic difference of function is enough to 
justify treating each consonant as phoncmically distinct from each 
vowel, and also for treating vowels and consonants as two differ¬ 
ent divisions of the phonemic system. Any other course, while it 
might reduce the inventory of phonemes, would greatly complicate 
our statements concerning the composition of utterances. Perhaps 
the most obvious of these structural differences between vowels 

248 Phonemic Systems 

and consonants is that every vowel is accompanied by a stress, 
and conversely, every stress is accompanied by a vowel. Another 
is that English vowels rarely occur in clusters, and with the 
exception of /o/ rarely or never at the ends of utterances. 

17.17 The problem of the complementary distribution of /s/ 
and /o/ seems trivial. There is no phonetic similarity, and every 
speaker of the language feels strongly that there is no close rela¬ 
tionship, a feeling which represents the speaker’s recognition of 
the structural cleavage between vowels and consonants just men¬ 
tioned. But there are other pairs where the problem is quite differ¬ 
ent. /yf shows a real phonetic similarity to /i/. Here, then, is a 
pair of sounds which are both phonetically similar and in com¬ 
plementary distribution, yet they are listed as two phonemes. 
This is necessary because of phonemic patterning within the lan¬ 
guage. There is a group of [i> or [y>likc sounds which contrasts 
with consonants: ycU [y61] : tell [t61], or Icy [16y] : let [16t]. Be¬ 
cause of this they must constitute a member of the consonantal 
system of English. There is another group of [i> or [y]-like sounds 
which contrast with vowels: bit [bit] : bet [Mt]. Because of this 
they must constitute a member of the vocalic system of English. 
We have two alternatives. The English phonemic system may be 
set up: 

Consonants /t y . . ./ or Consonants /t. . ./ 

Vowels /e i. . ./ Vowels /e . . ./ 

Ambivalent /i. . ./ 

The second alternative does reduce the inventory (/i/ stands in 
place of both /i/ and /y/), but at the expense of additional com¬ 
plexity of statement, both in the phonology and the morphopho¬ 
nemics. Moreover, though not a conclusive test by any means, the 
fact that the “average American” is not easily convinced of any 
identity of /y/ and /i/, even in spite of the common confusion of 
the two in traditional spelling, suggests that the first alternative 
is more nearly in accord with the structure of the language. The 
same problem arises with /w/ and /u/ and with /a/ and /o/. 

17.18 The problems of the last paragraph may give the im¬ 
pression of being trivial, but this is far from the case. In many 
languages there is no contrast between consonantal [y] and a 
vocalic [i] or the like. The overall phonologic structure of other 

Phonemic Systems 249 

languages may not require such clear distinction between con¬ 
sonantal and vocalic systems as does English. The American work¬ 
ing with such a language is easily led into wrong conclusions, as is 
the speaker of such a language who is working with English. The 
status of [i]- or [y]-like sounds — that is, the classification as 
consonant or vowel, or ambiguously as either, or perhaps as be¬ 
longing to a third category in the language — is one of the ques¬ 
tions which must be faced in making a thorough phonemic analysis 
of a language. Procedural techniques for making such a judgment 
cannot be presented here, but must obviously be based on attempt¬ 
ing to discover whether there is any such cleavage as that existing 
in English between consonants and vowels, what is the nature of 
that cleavage, and how each phoneme fits into the system. There 
seem to be languages in which the vowel-consonant dichotomy is 
much less significant than in English, perhaps even lacking. 

17.19 This suggests that the distinction between vowel and 
consonant is not phonetic, but phonemic. This is indeed true. The 
/r/ of many American dialects is phonetically more similar to the 
vowels than to the other consonants. However, /r/ patterns as a 
consonant. To avoid difficulty at this point, the terms vowel and 
consonant should properly be reserved for use in phonemics, and 
the terms vocoid (“vowel-like”) and non-vocoid should be used in 
phonetics. That serious damage is not done by the indiscriminate 
use of “vowel” and “consonant” is merely another instance of 
how closely phonetic and phonemic structural categories often 
parallel each other. Nevertheless, it is not essential that they do so, 
aud not infrequently there is a lack of conformity in detail be¬ 
tween the two. In these instances it is important to use care in the 
application of terms. 

17.20 The dichotomy between consonants and vowels is a basic 
one in English, but there is a still more fundamental division — 
that between vowels and consonants as a group, the four stresses 
and /+/ as a second, and the four pitches and three terminals as a 
third. Between these three sub-systems there is the same lack, or 
rather impossibility, of direct contrast as between consonants and 
vowels. Each of the three groups has such peculiarities of dis¬ 
tribution — within the phonologic structure of English utterances 
— as to clearly set it off from each of the others. In addition, this 
division within the phonemic system is paralleled by a division 

250 Phonemic Systems 

within the morphemic system which cuts the morpheme inventory 
into three groups. That is, some morphemes consist of (or are 
statable in terms of) consonants and vowels. These morphemes 
never contain phonemes of any other sub-system. Another set of 
morphemes is composed of stresses and / + /> and only of these 
elements. A third set consists of pitches and terminals, and only 
these. The three groups occupy quite different places in the gram¬ 
matical structure of English. Of course, there is no complete inde¬ 
pendence between the three sets of phonemes, nor between the 
three sets of morphemes, but the division is deep and basic. 

17.21 This division is, of course, a characteristic feature of 
English structure, as are all the lesser subdivisions. It is shared in 
all its essential outlines by some related languages, and by chance 
some others more remote may have something quite similar. But it 
is by no means a universal pattern. There is a very large number 
of languages in which the structure is quite different. Unfor¬ 
tunately, we as yet understand the full phonemic structure of so 
few languages that it is not possible to exhibit complete systems 
in contrast with the English system. However, some parts of the 
pattern are well established for many languages. In many, root 
morphemes consist of combinations of consonants, vowels, and 
pitches, and these parts of the phonemic system presumably are 
more closely related than comparable phonemes in English. We 
have reason to believe that in many such languages (“tone lan¬ 
guages”) there is another sub-system including phonemes com¬ 
parable to English terminals and associated with a grammatic 
sub-system comparable to English intonation. There are hints of 
other types of phonemic organization equally different from that 
of English in other ways. 

Whether a combination of three major phonemic sub-systems 
(as in English) occurs frequently in other languages or not, we do 
not yet know. I suspect that all languages will prove to have at 
least two, but I am very certain that the place of phonetically 
comparable elements in these sub-systems will vary widely from 
language to language. Forms of speech differ not merely in the 
inventory of the phonemes they contain, but even more funda¬ 
mentally in the ways in which they organize them into systems. 




Phonemic Problems 
in Language Learning 

18.1 No one ever has a command of more than a fraction — 
usually a very small fraction — of the vocabulary of a language. 
A normal individual learns new words daily, and at certain periods 
of his life makes vocabulary additions in great numbers. For an 
American student these include the latest slang, new words occa¬ 
sioned by technological advances, special terminology of the 
subjects he is studying, and miscellaneous words of all kinds 
learned as his contacts broaden. At the same time, words con¬ 
tinually pass out of his active vocabulary. Few people can recall 
the slang of a decade ago; and the forgetting of technical terms is 
the lament of students and professors alike. Vocabulary is a 
transient feature of any person’s command of a language, coming 
and going with comparative ease and rapidity. 

Any speaker of a language necessarily has a much more com¬ 
plete control of the grammar than of the vocabulary. To be sure, 
there may be syntactic constructions which he never uses, and 
some of these may not be completely clear when he hears or reads 
them. Many derivational patterns he may not recognize as such, 
even though he is quite familiar with words containing them. Some 
few features of the inflectional system may be relatively unfamiliar 


252 Phonemic Problems in Language Learning 

or completely strange to him. In short, few approach very closely 
to an absolute command of the morphology, but morphology is 
never as partially known or as transient as vocabulary. 

With the phonology, the demands are higher. A person who 
lisps _ that is, who uses an incorrect allophone of a single pho¬ 
neme, /s/ — is immediately spotted by a native speaker of English 
as deviant. Incorrect pronunciation of even a very rare phoneme 
or cluster can render speech conspicuously strange or even ob¬ 
jectionable. It can interpose a serious social barrier between the 
speaker and the members of the speech community. Of course, 
some deviations from the general usage are tolerated much more 
readily than others. While it is often possible to avoid the use of 
troublesome words or constructions, there is seldom any possibility 
of manipulating the conversation so as to avoid certain phonemes 
or phonemic combinations. 

A speaking knowledge of a language, therefore, requires very 
close to a one hundred percent control of the phonology and control 
of from fifty to ninety percent of the grammar, while one can fre¬ 
quently do a great deal with one percent or even less of the 

18.2 Command of the phonology is evidently a central prob¬ 
lem in learning to speak a language. It is, moreover, the point at 
which adult language learning is likely to be most unsatisfactory. 
Occasional persons do learn to speak a second language fluently 
and well without special procedures, but they are exceptional. 
Many immigrants living in America for decades still are unmis¬ 
takably foreign in their speech. Command of phonology shows little 
correlation with education; college professors and unskilled laborers 
have the same difficulties. Some even lose command of the mother 
tongue through long disuse without mastering their new language. 
If many years of hearing and speaking do not lead to an adequate 
command, it is evident that an adult must give some special atten¬ 
tion to the phonology in his study of a second language. 

Children seem to learn new languages comparatively rapidly and 
satisfactorily. One incidental reason is that they are less obsessed 
than adults with piling up a large vocabulary. Thus they tend to 
concentrate more on the patterns of phonology and morphology. 
The mast important factor is that they approach the new language 
with less deeply fixed habits of speech. At some time early in 

Phonemic Problems in Language Learning 253 

adolescence a point is reached where phonologic patterns become 
set; thereafter a new set of habits cannot easily be acquired without 
a sort of forced conformity to the old. No such point seems to be 
reached with regard to vocabulary, nor is there as great a degree 
of rigidity attained in morphology. The special problems of the 
adult in language learning are primarily phonologic. 

18.3 This chapter is entitled “Phonemic Problems in Language 
Learning,” and it is perhaps necessary to justify this title. We arc 
assuming that the objective is to learn to speak the language, not 
primarily to draw up a phonemic description. The two tasks are 
quite different. Of the billions of people who speak some language 
fluently and acceptably, only a few thousand have ever given any 
thought to the phonology of their language. Of those who have, 
many have entirely erroneous ideas on the subject. An under¬ 
standing of the phonemic system is certainly no prerequisite to a 
speaking command of any language. 

What is needed is the ability to operate the vocal apparatus 
without conscious effort so as to produce sound patterns character¬ 
istic of the language being learned. This is a matter of muscular 
habits, of recurring patterns of movement. The phonemic system 
of the linguists is basically a description of these patterns. If the 
mastering of a new phonology is to be attacked deliberately and 
systematically, there must be some descriptive statement as a basis 
for planning. This a phonemic analysis can provide. It is not, of 
course, all that is needed, since an adequate program must also 
be based on a sound theory of learning. 

For an adult to learn a language, it is not essential that he be a 
competent phonemicist. If his work is in charge of a teacher who 
has a thorough understanding of the problems, the necessary direc¬ 
tion can be provided from outside. Even in this case, however, some 
clear idea of the general nature of the problem is very helpful and 
may contribute both to the effectiveness and rapidity of his learn¬ 
ing. If such a teacher is not available — which is the more usual 
situation for many languages — then the student himself must 
provide the direction. Without this his efforts may be largely 
pointless, or, what is worse, they may be effective in establishing 
incorrect speech habits. As a very minimum, the student must 
understand the general nature of the problem, and be alert to 
every possible indication of inadequate pronunciation. Some spe- 

254 Phonemic Problems in Language Learning 

cific information on the special problems of the learning of the 
specific language with his specific background is a most desirable 
addition. An elementary understanding of linguistics may make 
the difference between success and failure. 

18.4 There are always two primary factors which must be 
taken into account. The first is that phonemic systems (or, more 
particularly, the phonological patterns which they describe) are 
incommensurable. No one can be effectively stated in terms of 
another. Pronunciation of an utterance in conformity with the 
habits of a different language will almost inevitably produce an 
unsatisfactory result. Hearing an utterance in terms of the patterns 
of another language may frequently result in failure to understand. 
The second factor is that the patterns of the mother tongue arc 
typically so deeply ingrained as to control the adult’s hearing of 
all sound, irrespective of the foreignness of its patterns. In learning 
a second language, the problems stem about equally from two 
sources, the new language and the old. 

While our concern here is with phonology, it may be pointed out 
in passing that the learning problems in connection with the 
morphology and syntax are much the same. Learning new patterns 
is only part of the process. Freeing oneself from old patterns is 
equally important, and sometimes more difficult. Information on 
the grammar both of English and of the new language is of help — 
though, of course, one must learn to speak from habitual patterns 
without any reference to such systematic formulations. 

18.5 Knowledge of the phonemic system of the second language 
is not enough. Textbooks of Hindi always give some description 
of the pronunciation, though few are couched in phonemic ter¬ 
minology. /p/ and /ph/ are often described as differing in that the 
latter has a “puff of breath” added, or something of the kind. This 
is true enough, and it is essential information. But taken by itself 
it can be exceedingly misleading. A student immediately perceives 
the problem as that of adding aspiration to an already familiar 
sound. Hindi /p/ is accordingly pronounced as English /p/, that 
is, usually as [ph]. To pronounce Hindi /ph/, an effort is made to 
add further aspiration. The results are quite various and almost 
universally unsatisfactory. Not infrequently something approach¬ 
ing [phoh] is heard. 

A more careful writer may state that Hindi /p/ is more or less 

Phonemic Problems in Language Learning 255 

similar to English p in spin, whereas /ph/ is like p in pin. This is; 
of no help to the average American because: (1) He does not under¬ 
stand the distinction intended, since he hears /p/ in spin and pin 
to be alike, as English phonemic patterns require. (2) He is unable 
to say the p of spin in such a context as Hindi /pol/ ‘bridge’ since 
his patterns are invariably to use [ph] in such an environment. He 
therefore confuses /pol/ with /phol/ ‘fruit.’ (3) The statement is 
illustrative — in a minor and comparatively innocent way — of a 
common confusion of sound and symbol, or of speech and writing, 
a confusion which is quite natural for him and which will continue 
to give him trouble unless it is actively combatted. 

18.6 Note that the instance mentioned in the last section is a 
phonemic, not a phonetic, problem. No new sound is involved. 
Both [p] and [ph] are quite familiar, as indeed are a number of 
other closely similar sounds. It is merely the usage that is new. In 
the one language, /p/ and /ph/ are allophoncs of a single phoneme; 
in the other, they are two contrasting phonemes. 

The phonologic problems in learning a second language are 
largely those of learning new uses for old sounds rather than learn¬ 
ing new sounds. Even in a language with clicks, popularly con¬ 
sidered the epitome of strangeness, there is little that is new. Every 
American knows several of these clicks and uses them in driving 
horses, in participating in a gossip session, or in amusing a baby. 
It is the use — that is, their place in the linguistic patterns, 
describable as their phonemic status — which is new and which 
creates a major difficulty. New sounds can cause difficulty, but 
they are never so large a part of the problem as the average student 

18.7 One desideratum, therefore, is a knowledge of English 
phonemics. That is one reason why four chapters are devoted to 
this topic in this book. These should give some understanding of 
the over-all patterns of English phonology. This, however, is not 
enough. To make effective use of this knowledge a student must 
know accurately how it applies to his own speech. He must know 
which of the thirty-six syllabic nuclei of English he habitually uses, 
and where. He must be able to identify them, and it is helpful if he 
can produce any of them at will. He must understand not only the 
inventory of phonemes, but also a good bit about the allophones. 
As we have pointed out, comparison of two phonemic systems by 

256 Phonemic Problems in Language Learning 

contrasting the lists of phonemes can be quite misleading. The 
point of contact between any two systems is phonetic. The prob¬ 
lem may be stated as that of disassembling the student’s speech 
patterns into the allophones, and of reassembling the allophones 
into new units, perhaps adding a few in the process. 

18.8 The second desideratum is a knowledge of the phonemic 
system of the language to be learned. This may be more difficult. 
There are four situations which need to be mentioned: 

1. There may be a relatively good analysis. Unfortunately, few 
languages have as yet had such analyses published. Some are pre¬ 
sented in a way that is particularly convenient for a learner; others 
are published primarily for the use of linguistic specialists and are 
presented in a very technical manner. In any case, if an analysis is 
available, the student who must plan his own work, or the teacher 
directing the study of others, has a relatively simple task. He can 
compare the statement directly with the known facts about the 
student’s native speech and his attempts at speaking the new 
language. This should suggest the points that will require special 
practice and the places where careful watch must be maintained 
to detect likely errors. 

We have seen that there is considerable variety in the analyses 
which have been proposed for the vowel system of English. Any 
one of them would be useful for a student learning English as a 
second language. (Some of them would be helpful only if he were 
learning that dialect to which the description applies.) This does 
not mean that they are all equally useful. Mast of them fail to dis¬ 
tinguish /\f from /a/, /i/, or /e/, and a native-like command of 
English phonology must include proper use of /*/. However, errors 
at this point are less serious than those involving most other pho¬ 
nemes. Indeed, the fact that /»/ could so long be overlooked is 
largely explained by the fact that, while it is certainly phonemic, 
it is set off from other phonemes by contrasts which are few in 
number and relatively unimportant in function. 

For most foreigners learning English, there is a pedagogic value 
in those analyses which emphasize the diphthongal nature of Eng¬ 
lish “long vowels.” This is one of the most characteristic features 
of the English phonemic system. A common error of many for¬ 
eigners in learning English is to fail to make many of these diph¬ 
thongal, using instead the pure vowels to which past experience 

Phonemic Problems in Language Learning 257 

has habituated them. Conversely, a common feature of an “Eng¬ 
lish accent” in many languages is the use of these diphthongal 
nuclei in place of pure long vowels. Since this is so, there is a slight 
but real advantage in making the contrast clear by an obviously 
diphthongal notation for English. 

In many other languages such divergent analyses are possible. 
Any type of analysis is usable to the extent that it actually states 
the facts of the language. When there is a choice in the interpre¬ 
tation of the data, that analysis is to be preferred which most 
clearly exhibits the differences between the language to be learned 
and the learner’s mother tongue. 

18.9 2. There may be a good but incomplete analysis. One of the 
common faults of descriptions, from the point of view of their 
usefulness in planning a language-learning program, lies in their 
failure to describe pitch, stress, and rhythm, and particularly 
transitions and terminals. In many languages, adequate descrip¬ 
tions exist of consonants and vowels, and nothing whatever of 
other parts of the system. The danger is that such analyses, 
actually useful as far as they go, may l>e worse than useless, by 
reinforcing the tendency of students and teachers alike to neglect 
these other facets of phonology. 

In general, patterns of intonation and rhythm are the sources 
of the greatest difficulties to the average American, though many 
of them never realize it. They must receive considerable intense 
direct attention. This must come early in the program, before 
speech habits involving English-like intonation are entrenched. 
First priority should be given to these aspects before every other 
feature of the language. If the available sources do not describe 
them in a useful way, the student or his teacher must work them 


Another type of incompleteness that reduces the usefulness of 
an otherwise excellent phonemic description is the omission of 
adequate phonetic descriptions of the allophones. A mere list of 
the phonemes will, of course, provide a useful warning against 
various errors that a student can make. But it is not enough to 
guide a student in his work. It will not make clear the relationships 
between the two phonemic systems with which he must deal. They 
can be compared only on the phonetic level. If an adequate state¬ 
ment is not given, it must be more or less supplied by identifying 

258 Phonemic Problems in Language Learning 

the phonemes in the speech of the native informant and observing 
their phonetic character. 

18.10 3. There may be a poor analysis. The wise course is to 
reject it entirely and proceed as if there were none. The problem 
is to recognize it as poor. The only final recourse is to check the 
statement against the speech of a native informant. The best 
linguist may produce a statement of phonology which is for the 
learner's purpose seriously incorrect. It may be based on a different 
dialect. It may even have been based on observation of an in¬ 
formant with a speech impediment (This has happened!). Even if 
these are remote possibilities, they cannot be wholly discounted. 
The speech of the native informant is always the final authority. 

There is usually some internal evidence of poor analysis of 
phonology. The following, particularly several of them together, 
may be considered as presumptive evidence of untrustworthiness: 

(a) Some state pronunciations in terms of English equivalents 
with little or no qualification. Even when competently done (a rare 
case!) this can be misleading. “Bata . . . does sound very much 
like barter .” This is not very bad (nor good enough) if you pro¬ 
nounce the latter /b&Hto/, as presumably the author does, but 
very bad if you say /b&rtor/ or /baurtar/ as most Americans do. 
The facile statement that "vowels arc as in Italian” is more com¬ 
monly made by people who speak no Italian than by those who 
do: it is almost always meaningless. 

(b) Some direct students to overlook features. "Some con¬ 
sonants are aspirated, and should then be written with a h fol¬ 
lowing: e.g., dhunhu ‘a small hill.’ As, however, the aspirate is not 
always distinguished by Europeans, we have decided to disregard 
it, except where its omission might cause confusion.” 

(c) Some pessimistically state that a proper pronunciation is 
impossible to acquire. Presumably the author himself failed, and 
may have failed likewise at other points. "TL — wholly inde¬ 
scribable, almost unattainable, and very seldom used.” 

(d) Some make free use of impressionistic and largely meaning¬ 
less terms. "Make the vowel as full and rich as you like.” 

(e) Some express linguistic features as the result of racial or 
other irrelevant attributes. "His flighty and pleasure-loving dis¬ 
position is the most noticeable of his characteristics, and shows 
itself particularly in his language.” 

Phonemic Problems in Language Learning 259 

(f) Some seem particularly concerned with sounds that are not 
in the language being described. “The consonants /, q, v, x, z, do 
not exist.” 

(g) Some descriptions utterly confuse spelling with pronuncia¬ 
tion. “This restricted orthography accounts for some of the 
multiple and confusing letter sounds which appear in the following 

(h) Some seem confused by the native speaker’s seeming per¬ 
versity: . it seems impossible for a native, even in spelling, to 

give g the simple hard sound of g in go — it always has the nasal 

(i) Some confuse phonemics and phonetics. “Many phoneticians 
claim that ch and j are compound sounds. They were probably led 
to this error because the sounds are not heard in French or German, 
and they were led to analyze them as compounds of tsh and dzh in 
order to explain them. All nations that have them have heard them 
as simple sounds. . . .” The last statement might be paraphrased 
“In all languages where they occur as phonemes they are pho¬ 

(j) Some state the pronunciation of the present day language 
in terms of deviations from a known or reconstructed parent 
language. “This letter is pronounced by ... as a in most lan¬ 
guages, or as ah or the a in father in English. It has never been 
doubted that this was the original sound. . . .” 

(k) Some are patently contradictory. “. . . there are no diph¬ 
thongs. . . . The following five English words give pretty nearly 
the sounds of the five . . . vowels, a, e, i, o, u, far fail feel foal 

(l) Some suggest ridiculous gymnastics. “. . . it is best de¬ 
scribed as an attempt to pronounce feed with the front part of the 
mouth and food with the back part of the mouth.” 

Note however: Statements that do not include the terras 
“phoneme,” “allophone,” etc., or statements not in the usual 
format employed by descriptive linguists, are not necessarily to be 
rejected. Some excellent descriptions have been produced by 
writers with little technical training, or with training received prior 
to the present development of phonemic terminology. Some others 
have been written by phoneme-conscious linguists in deliberately 
non-technical language. Conversely, some authors have used tech- 

260 Phonemic Problems in Language Learning 

nical terminology with no adequate understanding of their 

18.11 4. There may be no analysis whatever. In this case the 
student or the teacher must make one. This does not mean that 
he must draw it up in the finished form in which a professional 
linguist would publish it, though this might be highly desirable 
for other reasons. But he must be constantly on the lookout for 
any significant contrast or lack of contrast, and be prepared to 
draw as much as he can from any hints in his informant’s speech, 
his own errors, his informant’s corrections, and the reactions of 
others. His analysis may be inchoate, but it must be adequate to 
direct his attention to the trouble spots. It will be helpful if the 
student can produce a phonemic script in which he can keep his 
notes and prepare his drill materials. 

The remainder of this chapter will be devoted to some hints on 
this matter supplemental to the techniques described in Chap¬ 
ter 13. They will be stated primarily from the point of view of the 
needs of a language student, but much of what is said will apply 
also to a technical linguistic analysis. 

18.12 In 13.2, four types of errors were listed as commonly 
occurring in preliminary transcriptions made for phonemic anal¬ 
ysis. They arc also characteristic of the student’s first gropings 
after a new phonemic system. Not all are equally serious. In gen¬ 
eral, a moderate amount of over-differentiation is not disastrous. 
It will add some unnecessary complications to the morphology and 
increase the load that the student must master. Excessive over¬ 
differentiation may leave the student hopelessly enmeshed in de¬ 
tails that might have been avoided by correct classing of the 
allophones into more manageable units. 

Under-differentiation is much more serious. It may render speech 
unintelligible. Unfortunately for the student, it does so only in 
certain instances. He can usually get away with poor pronuncia¬ 
tion long enough to become thoroughly established in habits which 
will lead to trouble later. Even when misunderstandings do not 
arise, the result can be quite obnoxious to native speakers. Consider, 
for example, your own reactions to the common failure of certain 
foreigners to distinguish /0/ from /s/ or /t/, and /<5/ from /z/ or 
/d/. The occasional total misunderstanding is minor compared 
with the constant irritation that such a pronunciation occasions. 

Phonemic Problems in Language Learning 261 

No systematic procedure will discover and correct under-differ¬ 
entiation. But there are certain situations which should cause 
suspicion and lead to careful rc-examination of the data. As in so 
many instances, the only definite answer must come from careful 
rechecking with a cooperative native informant. 

18.13 One indication of possible under-differentiation is exces¬ 
sive numbers of apparent homophones, that is, words which appear 
to be the same in pronunciation. Every pair of homophones should 
be checked with the informant to determine whether he considers 
them as actually alike. If he reports them as different, or especially 
if a second informant is able to identify correctly the words which 
the first informant is asked to pronounce, there is under-differen¬ 
tiation. The task is then to discover what is the feature that has 
been overlooked or confused with another. 

There is an accessory value in this. Many of the pairs will prove 
to be minimal. These constitute exceptionally valuable drill mate¬ 
rial, since errors cannot be so easily overlooked. Moreover, if they 
were not distinguished at first, they will be minimal pairs for 
precisely those contrasts on which drill is most needed. 

18.14 Another means of discovering under-differentiation is to 
make careful note of utterances which are misunderstood. If cer¬ 
tain items keep reappearing, it is indicative of some trouble in the 
pronunciation of these particular words. If many of these contain 
the same “sound,” it may well be that the trouble is in that place. 
A careful check should be made, taking note of the words or 
“sounds” that seem to be misunderstood. Errors are important 
data in correcting pronunciation as well as in extending analysis. 
They should be noted with both uses in mind. 

18.15 In some cases, apparent irregularities in morphopho- 
nemic changes may suggest places where the phonology is suspect. 
In Loma (Liberia) word initial phonemes are changed in certain 
contexts which cannot be defined here: 

[p61e] becomes O&e] and so all 


become [v] before [i e e a] 




[w] before [u o o] 













[bflfj becomes 

[vili], and so many 

[b] become [v] before [i e e a] 

262 Phonemic Problems in Language Learning 

£bu] [wil] [b] [w] before [u o o] 

£b&f&] remains [b&fii] and so many [b] remain [b] in any con¬ 

The system is quite regular. (The tabulation is fragmentary; the 
regularity would be more evident if all the possible initial con¬ 
sonants were considered.) It is therefore suspicious that [b] should 
show two contrasting patterns, one like [p] (as [d] is like [t]), and 
one of no change. Investigation will show that the transcription 
£baf&] is incorrect. It should be [6af&], with a phonemically differ¬ 
ent initial, an implosive voiced bilabial stop contrasting with the 
simple voiced bilabial stop of the other examples. With this cor¬ 
rection, the transcription is now phonemic; [ ] can properly be 
replaced by / /. The patterns of initial consonant change are found 
to be quite regular. Of course some languages have morpho- 
phonemic patterns that are much less symmetrical, so that ap¬ 
parent irregularities are not necessarily evidence for such a 

18.16 Whatever consideration calls attention to a possible 
under-differentiation, the next step is to make new recordings. 
Then these must be carefully checked to be sure that the distinc¬ 
tions now recorded are actually phonemic. It is very easy to 
imagine you hear a difference when you strongly suspect that there 
should be one. You must be sure that the contrast you hear is real 
and that it is significant. 

18.17 At the first approach to a new phonemic system, stu¬ 
dents are occasionally troubled by apparent random variability 
in the pronunciation of certain words. Sometimes this is due to a 
vacillation between two alternative pronunciations of one mor¬ 
pheme, similar to that between /ruf/ and /ruwf/ for some Amer¬ 
icans. This is not a phonemic problem, as is evident from such 
pairs as look /ltik/ : Luke /ltlwk/, in which no such variation is 
observed. Other instances are quite obviously of a different type. 

One clement in learning to understand a spoken language is to 
learn to hear every sound as an instance of one of the phonemes 
of the language. There is, of course, a certain amount of random 
variation within each allophone, as was pointed out in 12.8. Most 
instances of a given allophone are more or less similar to a mean 
value for all instances, and also rather more distant from the mean 

Phonemic Problems in Language Learning 263 

values of any other phonemes. In other words, each allophone has 
a characteristic range of variation. 

Let us consider a large population of instances of each of two 
phonemes, /X/ and /Y/, in some simple environment in which 
they contrast. The distribution in regard to some single measur¬ 
able factor might be something as follows: 

This large population is perhaps the sum of past experience which 
has set the phonemic pattern of the individual hearer. This will 
lead him to interpret any sound near the mean value of /X/ as an 
instance of /X/, and an}' sound near the mean value of /Y/ as an 
instance of /Y/. A sound halfway between the mean values will 
be interpreted more or less randomly as either /X/ or /Y/. Only 
if the distance from both mean values is great will it be heard as 
a “strange sound.” Of course, such intermediate sounds are very 
rare in the language from which the past experience was drawn. 
That fact is represented in the diagram by the dip in the frequency 
curve between the two humps. 

Suppose this hearer starts to study a language in which there 
is no phoneme comparable to either /X/ or /Y/, but one, /Z/ 
whose mean value is intermediate between those of /X/ and /Y/. 
The diagram below represents all three of these. Until the learner 

assimilates the new pattern and establishes a new set of norms by 
which to classify the sounds he hears, each instance of /Z/ will 
sound to him as either /X/ or /Y/. Which reaction occurs will be 
partly the result of randomness of his responses (particularly when 

264 Phonemic Problems in Language Learning 

an instance of /Z/ is near its mean value), and partly of the ran¬ 
domness of the pronunciation of /Z/ (particularly when the 
instance of /Z/ is less close to its mean value). Successive repeti¬ 
tions of the same word may sound quite different (one with /X/ 
and one with /Y/) to the student, but quite alike (both with /Z/) 
to the native speaker. The one hears them as more different than 
they actually are, the other as less different. 

18.18 By this time some such experiment as the following has 
undoubtedly been used in your class. Some non-English sound, 
say ft], is pronounced and the class is asked to identify it. Some 
will call it [t] and some [k]. Very few will report it as wholly 
strange, though a few will call it a “strange [t] M or a “peculiar 
[k]” or something of the kind. Most Americans will assign ft] to 
one or the other of their English phonemes, until they have been 
trained to recognize it as a distinct entity. Some will be reasonably 
consistent in calling it ft], others more or less consistent in identi¬ 
fying it as [k]. Some will hear it sometimes one way, sometimes the 
other. The proportions will depend in part on the precise degree of 
retroflexion used. The difference between different students in 
their reactions is the result of difference in past experiences. 

Vacillation in hearing what the informant insists is identical 
should suggest this type of explanation. The lack of agreement 
between observers, if two or more are present, should raise the 
same possibility. Sometimes such confusion can be cleared up by 
tabulation. Sometimes it may cause considerable difficulty in the 
distributional patterns of other phonemes. Usually the problems 
arc mast easily eliminated by closer attention to phonetics and by 
rechecking with the informant. 

18.19 What has been said here must not be taken to imply 
that phonemic analysis, however thorough and competent it may 
be, can either assure — or, even with good fortune, produce — an 
adequate command of the phonology of a language. It cannot. It 
can only provide a basis for planning the work and illuminating 
the problems which arise. This is, however, an essential service. 
Together with training in phonetics, close attention to phonemics 
may enable a student to make an excellent first approximation 
quickly and easily. Two problems remain: to achieve a pronuncia¬ 
tion which is acceptable in the last detail, and to make it all 
automatic. The ultimate polishing can only be had by close atten- 

Phonemic Problems in Language Learning 265 

tion to the speech of an informant, to mimicry of every detectable 
mannerism, and to constant drill. These things are the real essen¬ 
tials. But without direction they can become quite as profitless as 
the}' are tedious. There is no easy road to language learning. How¬ 
ever, an intelligent approach may both expedite progress and 
insure a better result. 

chapter | 




The Process 
of Communication 

19.1 Language, as we have been examining it, is a complex of 
structures of various kinds. The analysis of a language must pro¬ 
ceed by separating out the various parts, but a full understanding 
of language cannot be gotten if they are left as detached details 
unrelated to one another. The various elements are of significance 
and interest primarily because they fit together into one integrated 
system which people use in communication. This function of 
language provides a framework within which language can be 
looked at more or less as a whole. 

Such an approach to language can be fruitful because com¬ 
munication is a much broader process than language. It includes 
a number of phenomena which, though showing basic similarities 
to language, are very much simpler. Starting from some of the most 
elementary forms of communication it is possible to develop cer¬ 
tain principles which will be applicable to language. In recent years 
a whole new branch of science, communication theory, has arisen 
in this way. Though the full implications for linguistics are just 
beginning to emerge, it is already quite evident that some powerful 
new insights are likely. 


The Process of Communication 267 

19.2 There are certain elements which can be listed as essential 
to any communication process, including language. The following 
are of some interest to us: 

1. A code, an arbitrary, prearranged set of signals. A language is 
merely one special variety of code; and the science of linguistics 
deals, in its strictest delimitation, only with this aspect of com¬ 

2. A channel, some medium by which the signals of the code are 
conveyed. This may be audible sound in the range from about 
100 cps. to perhaps 7000 cps. (as in the case of speech), a band of 
radio frequencies (compare the now familiar use of the term 
“channel” with reference to television), light, electrical impulses 
carried on a wire, or mechanical devices of various kinds. 

3. The process of encoding, by which certain signals in the code 
are selected and put into the channel. The selection is typically 
done in response to some outside condition — that is, one com¬ 
municates what is observed. The way in which this is to be done 
is the meaning of the code, and is prearranged. 

4. An encoder, the person or device which performs the process 
of encoding. 

5. The process of decoding, by which the signals are identified 
and a course of action is affected by them. 

6. A decoder, the person or device by which the process of decod¬ 
ing is performed, and whose course of action is thereby affected. 

It is worth pointing out that either the encoder or the decoder or 
both can be either persons or machines. A traffic light is part of a 
communication system; it does not matter whether the signals are 
selected by a policeman or a mechanical device. An automatic 
temperature control system is a communication system in which 
the encoder is a device in the living room and the decoder another 
device in the basement. The human factor may be essential in 
language, but not in communication generally. 

19.3 The significance of these elements can be seen in a very 
simple and familiar case — the case of Longfellow’s account of a 
central event in his poem “Paul Revere’s Ride.” The code was an 
extremely simple one: “One, if by land; and two, if by sea.” This 
was arranged in advance between the two principles: the encoder, 
the friend who remained in Boston, and the decoder, Paul Revere, 
who crossed the river to Charlestown. The process of encoding was 

268 The Process of Communication 

the selection of the proper signal (two lanterns) and hanging them 
in the belfry of Old North Church. When Paul Revere saw them, 
he mounted his horse and rode off to Lexington: the process of 
decoding was not merely the identification of the signal; it deter¬ 
mined which of two previously planned courses of action he was 
to follow. 

In a code such as this there are certain very evident limitations. 
It could transmit only a very limited amount of information. No 
hint could be given about the size of the force, the time of de¬ 
parture, details of the route, or any other facts, no matter how 
important, which the friend might have picked up in his wander¬ 
ings through Boston. The two men could, of course, have agreed 
in advance on a code which would have permitted them to transmit 
as much detail as they cared. But it would have been wasteful to 
agree on a very elaborate code unless they had foreseen some real 
value in further details. Presumably they believed that this simple 
code would be adequate for their purposes, without involving them 
in any unnecessary complexity or danger. A code can be designed 
to transmit any amount of intelligence that may be desired, but 
there are always certain limitations on what any given code can 

19.4 The notion of the capacity of a code, and the closely 
related one of the actual amount of information transmitted, have 
proved to be very useful, once proper definitions are set up to 
allow of precise measurement. A measure of the amount of informa¬ 
tion which a code can convey is obviouslj' related to the number of 
alternative signals. Paul Revere’s code is at the minimum, with 
two signals. The amount of information increases as the number of 
alternatives increases. — These statements are not so much ob¬ 
vious facts about codes as they arc the beginning of a technical 
definition of information. By this definition information is by no 
means the same as in any popular definition of the term. 

Information is measured in units called binits. (“Binit” is pre¬ 
ferred by many linguists; communication engineers usually use 
“bit.”) By definition, a code with two alternative signals, both 
equally likely, has a capacity of one binit per use. A code with four 
alternatives is defined as having a capacity of two binits per use; 
one with eight alternatives, a capacity of three binits. That is, 
the capacity in binits of a code of this type is the logarithm to 

The Process of Communication 269 

the base two of the number of alternative signals in the code. This 
mode of definition is made desirable by certain mathematical con¬ 
siderations which are beyond the scope of this present discussion, 
as are likewise most of the direct mathematical consequences of the 

19.5 For the benefit of those students not conversant with the 
mathematics involved, the following rough interpretation of loga¬ 
rithm to the base two may be given: 

If we start with 1 and multiply by 2 repeatedly until we reach 
the number of signals, the number of times we multiply will be the 
Log* of the number of signals: 




Such a definition is not adequate for our purposes, since we may 
need to consider a code with some other number of signals, say 3. 
We might expect that Log* 3 would be somewhere between 
Logs 2 = 1 and Log? 4 = 2. We cannot easily imagine multiplying 
by 2 one and a fraction times, nor can we readily see what the 
fraction would be. The value is a little below 1.585. (Most loga¬ 
rithms will not come out even, no matter how many decimal places 
arc calculated.) Just to give some idea of what such logarithms are 
like, we give here the following short table: 


- 1 


1 X 2 

= 2 

Logs 2 


= 4 

Log* 4 

1 X 2 X 2 X 2 

= 8 

Log 2 8 


= 16 

Log* 16 


= 128 

Ix)g* 128 

Log* 1 


Log2 9 






























19.6 One crucial aspect of this definition of information can be 
made clear by another very simple communication system. Con¬ 
sider a store with a burglar alarm. This is a part of a communica- 

270 The Process of Communication 

tion system in which the encoder is a mechanical or electrical 
device, and the code includes only two signals: silence or alarm. 
Now consider the policeman whose beat passes the store, and 
whose responsibilities include acting as the decoder in this system. 
Night after night he goes by, decodes the signal of silence, and 
responds by continuing on his beat hardly conscious of the store 
or of the communication system in which he is functioning. But 
finally, one night the alarm rings; his usual nightly routine is sud¬ 
denly and drastically changed. 

Since we have defined the decoding process in terms of deter¬ 
mination of a course of action, we are forced to the conclusion that 
the communication process is markedly different when the alarm 
rings from when it does not. Reflection will show that the difference 
rests in the unexpectedness of one signal, compared with the fre¬ 
quent and expected occurrence of the other. We express this by 
saying that more information is conveyed by the alarm than by 
the silence. 

19.7 We have defined the capacity of a code of two signals as 
equal to one binit. This was predicated on the signals being equally 
likely. In a system such as we have just described, the amount of 
information in the alarm is much greater than one binit; that in 
the silence much less. An exact mathematical definition is possible: 
The amount of information in any signal is the logarithm to the 
base two of the reciprocal of the probability of that signal. That is: 
I - Logs 1/p. 

This formula may require some explanation for some students. 
The probability of a signal may be considered as the proportion of 
all the signals which are instances of the signal concerned. For 
example, suppose that in a given code there are two signals. We 
observe a large number of messages and find that on the average 
signal A occurs once to every fifteen occurrences of signal B. We 
therefore estimate the probability of signal A as 1/16 or .0625, 
and that of signal B as 15/16 or .9475. Probabilities will always be 
less than 1, and the sum of the probabilities of all the signals in any 
code will always be exactly 1. The reciprocal is merely a fraction 
turned upside-down. The reciprocals of the probabilities of the 
two signals are therefore 16/1 or 16.0 for A and 16/15 or 1.067 
for B. The amount of information I for signal A is therefore 
Log* 1/(1/16) or 4.00 binits. For signal B it is 0.093 binits. Signal B 

The Process of Communication 271 

is 15 times as frequent as A, but carries only 1/43 as much in¬ 

From the amount of information in each signal we may calculate 
the average amount of information per signal. Using the same 
example — 

1 occurrence of A, each carrying 4.00 binits = 4.00 

15 occurrences of B, each carrying .09 binits - 1.39 

16 occurrences carrying a total of 5.39 

Divide by 16 to get an average of .34 binits 

The details of this calculation are much less important than the 
following principle which it illustrates: The full capacity of a code 
is realized over a period of time only if all the signals have equal 
probabilities of occurring. In this instance, the unequal frequencies 
of the two signals reduce the efficiency of the code to about one 
third, since if they were equally likely, the capacity of the code 
would be 1.00 binit. 

19.8 Suppose we set up a code in which there are two signals 
A and B, both equally probable, which can be sent at the rate of 
one signal per second. By our definition this code has a capacity of 
one binit of information per second. Suppose we send for four 
seconds. Any of the following messages might occur, all with equal 

Sixteen signals of equal probabilities represent a capacity of 
4 binits. Since each such message requires four seconds for trans¬ 
mission, this works out to a capacity of one binit per second. This 
figure checks with our first calculation. (Incidentally, to make this 
sort of check possible is one reason that information is measured 
in terms of logarithms in the way we have described.) 

Now suppose it is agreed that each signal will be repeated ex¬ 
actly twice. If no other change is made in the system, any of the 
following messages can be sent in four seconds: 


Our decision to repeat has reduced the capacity of the code to 
only 2 binits in four seconds, or to half a binit per second. That 
is, it has reduced the efficiency of the code by one half. 

272 The Process of Communication 

The unused capacity is the result of our decision to repeat, and 
so may be labeled as redundancy. Here again, this term is not 
necessarily used in any sense in which it is popularly used, as we 
shall see. It is a technical term in information theory, and requires 
a technical definition in terms that will fit with the definition of 
information. We define redundancy as the difference between the 
theoretical capacity of any code and the average amount of in¬ 
formation conveyed; it is expressed as a percentage of the total 
capacity. Thus, our decision to repeat each signal twice introduces 
a redundancy of 50 percent. 

Redundancy is not synonymous with repetition, nor is it neces¬ 
sarily the result of repetition. In the last section we calculated the 
average amount of information in a signal for which the prob¬ 
abilities are 1/10 and 15/16 to be .34 binit. Since the average 
amount of information would be 1 binit if the probabilities were 
equal ( 1/2 and 1 / 2 ), the disparity in the probabilities introduces 
an average redundancy of 66 percent. 

19.9 In the example of the last section we calculated the re¬ 
dundancy by listing all the possible messages (16) and all the 
messages (4) which could be used under the special restriction 
imposed, and computed the amount of information in each case. 
This is quite feasible in such an example, but under some condi¬ 
tions would be very unwieldy or even totally impossible. There are, 
for example, 205,962,976 theoretically possible combinations of 
five phonemes, taken from the English stock of 46 phonemes, if no 
special restrictions are imposed. It is easy enough to compute the 
total information which might be carried. Log 2 46 equals 5.52; 
there are therefore 5.52 binits of information in each occurrence 
of one phoneme, and 27.62 binits for a sequence of five phonemes. 
But to list all the sequences of five phonemes which are actually 
used in English would be exceedingly laborious, and would not 
yield a very accurate picture because of the unequal probabilities 
of the various combinations. Fortunately there is another way of 
approaching the problem. 

Consider again the example of the last section. The first signal 
can be either A or B, and the probabilities are equal. This first 
signal, therefore, carries one binit of information. Suppose it turns 
out to be A; consider the probabilities for the second signal. Here 
it is no longer equally likely that it will be either A or B, since we 

The Process of Communication 273 

had agreed that we would repeat each signal twice. It is therefore 
a practical certainty that it will be A. The amount of information 
carried by this signal is therefore zero binits. The third signal can 
be either A or B, so this signal carries one binit of information. 
Since the fourth signal is also a repetition of the third, it is known 
in advance, and so no information is conveyed. We may in this 
manner compute the total information in each signal separately, 
and add these figures to get the total amount of information in 
the message as a whole. The results of the two methods, provided 
each calculation is done thoroughly, are the same. If the com¬ 
plexities are such that it is not possible to do the calculations thor¬ 
oughly, it may nevertheless be possible to do them with sufficient 
completeness to provide a usable approximation to the correct 

19.10 Using this method, let us examine the redundancy in 
some language materials. Just for convenience, we shall use written 
English for our first example; spoken language could be examined 
in much the same way. 

An English sentence can begin with any letter whatever. How¬ 
ever, there are wide variations in the frequencies of various letters. 
Sentences like Xerxes was a Greek general, are quite possible, but 
very rare. Sentences beginning with T are very common; seven 
of them occur in section 19.9, in which the total number of sen¬ 
tences is only nineteen. A count of a moderate sample of sentences 
gave the following frequencies of initial letters: 





















































If all twenty-six letters had equal frequency, that frequency would 
be .038. Nine of the letters exceed this average. T is 6.1 times as 
frequent as the average. On the other hand, seventeen letters are 
less frequent than the average, most of them very much less fre¬ 
quent. With such disparity, we would expect appreciable re¬ 
dundancy. The average amount of information in the first letter of 
an English sentence is about 3.10 binits. If all twenty-six letters 

274 The Process of Communication 

were equally probable, the total information would be 4.70 binits. 
The average redundancy, considering only the first letter in a 
message, is therefore about 34 percent. 

Suppose now that we examine the second letter in each sentence. 
Let us take the commonest case, that where the sentence begins 
with T. There is very little freedom of occurrence. As the second 
letter, x and l do not occur at all, or perhaps very rarely in sen¬ 
tences starting with non-English words, as Tlalocan is a Mexican 
anthropological journal. Overwhelmingly the commonest is h, hav¬ 
ing a frequency of about 88 percent. Next most probable is o, with 
a frequency in the neighborhood of 6 percent, while a, e, i, u, w, r 
are occasional and make up most of the remaining 6 percent, and a 
few others, though rare, arc not wholly unexpected. ( Tschaikowsky 
was a musician, etc.) The redundancy figures to about 83 percent. 

But this is by no means the extreme. Consider the sentences 
which begin with Q. Almost every one of them has u as the second 
letter. Exceptions would include only such rare sentences as 
Qaraqalpaq is a Turkic language. The redundancy here approaches 
very closely to 100 percent. 

The process can be continued as far as we desire. After Th, e has 
a frequency of about 83 percent, i about 8 percent, a about 3 per¬ 
cent, and so on. Several of the letters will not be found at all. The 
redundancy figures to about 77 percent. One more step: If the 
sentence begins with The, the next character is likely to be space 
(53 percent), y (18 percent), n (14 percent), r (12 percent). Clearly, 
this is because the, they, then, and there are not only very common 
words in English, but more likely to occur at the beginnings of 
sentences than elsewhere. 

19.11 We have here our first indication of a source of the high 
redundancy of written English. In addition to variations in the 
frequencies of letters as such, and to certain restrictions on 
sequences of letters as such (q is generally followed by u ), there are 
considerable restrictions imposed by the fact that letters are used 
to spell words, and that words as such have widely differing fre¬ 
quencies and places of occurrence. 

In any code comparable to written English, there are various 
levels of organization. Each of these imposes on messages certain 
restrictions w’hich are reflected in the redundancy of the language 
as a code. These restrictions arise simply because such codes have 

The Process of Communication 275 

structure. Structure is merely a set of limitations on freedom of 
occurrence and hence inevitably produces redundancy. 

19.12 Before we go on, it would be well to get away from 
written language, with its special problems yet to be discussed 
(see Chapters 21, 22, and 23), even though it does afford con¬ 
venient material for analysis and presentation. The situation is 
not vastly different with spoken English. We would find for ex¬ 
ample that /8/ is a relatively common phoneme at the beginning 
of English sentences (about 18 percent); /o/ does not occur at all 
in this position. After /$/, /o/ would be found to be very com¬ 
mon, with a frequency in the neighborhood of 50 percent. After 
/&>/, / + / would be very common. This would be merely a reflec¬ 
tion of the fact that the morpheme /So/ the is very common in this 
position in English sentences. The method of procedure is not 
different from that which might be used with written language, 
except that the material for analysis would have to be transcribed 
first, and it would be more necessary to specify what dialect is 
represented in the data. 

19.13 We may list the following as particularly interesting 
sources of redundancy in spoken English. The list is not exhaustive: 

1. variation in frequency of phonemes. It is characteristic 
of language that all elements vary widely in frequency. The kind 
of distribution illustrated above for English initial letters, or for 
French verb classes graphed in Chapter 9, may be expected with 
any other kind of data. For example, in my own speech /i i o e/ 
are comparatively common; /o/ is quite rare. Among the con¬ 
sonants, /t n s/ are much commoner than /0 £/• /4/ is much less 
frequent than the other pitches. 

speech /i/ is very frequent, but the sequence /iw/ is very rare, and 
/iy/ or /jh/ are not used at all. By contrast /ow/ is much more 
common than /of without a glide, /iq/ is comparatively common, 
but /iyq/ does not occur, /nk/ is rare, but /qk/ is common. With 
every vowel there occurs one and only one stress. All such limita¬ 
tions — and a very large number can be listed — increase re¬ 
dundancy by making individual phonemes more probable in cer¬ 
tain statable environments and less probable in others. 

3. non-use of possible morphemes. Such sequences as /Oat/ 
or /siyg/ are quite possible under the restrictions imposed by 

276 The Process of Communication 

English phonologic patterns and could easily be used as mor¬ 
phemes. The fact is, however, that, for largely fortuitous reasons, 
they are not. Considerably more than half of the English mor¬ 
pheme shapes of this order of size are not used. Such complete 
non-use of certain signal patterns must contribute to redundancy. 
But it is worth noting that this reserve of possible morphemes is 
of vital importance if the language is to alter its vocabulary to 
meet new needs, as every language must. 

4. variation in FREQUENCY of MORPHEMES. The striking illus¬ 
tration of this is found in the English /<5/- This probably occurs in 
fewer morphemes than any other English consonant (except pos¬ 
sibly /If). However, many of them, noticeably/So/, /Sey/, /Sat/, 
/Sot/, /Sen/ are among the commonest morphemes in the lan¬ 
guage. Largely as a result of this, /'$/ is a relatively common 
phoneme. Moreover, it is noticeably more common at the begin¬ 
nings of words than elsewhere, in spite of the fact that it occurs at 
the beginning in only a few frequently occurring words. The same 
circumstance, of course, explains the very high frequency of /o/ 
after /S/. 

restriction necessarily increases redundancy. But restrictions on 
distribution are essential if the language is to have structure. A 
language in which morphemes could be placed in any order at will 
would be completely inoperative. Such a language would be unable 
to express any relationships between morphemes, and hence an 
utterance would be nothing more than a list of vocabulary items, 
about as illuminating as a random sample from a dictionary. 


While such a sentence as The green absolute signals the ineffable 
hypotenuse, is grammatically quite acceptable (i.e., it follows a 
common syntactic pattern), it is quite unlikely to be said, since 
no context is imaginable. But again, it is essential to the flexibility 
of language that utterances which cannot be foreseen in advance 
to be useful are possible. What would have been thought a century 
ago of a sentence like Light is both particle and wave.? All the 
vocabulary was available then. So was the syntactic pattern. But 
the sentences would probably have been about as intelligible as 
that cited just before it. If English could not have allowed an 
utterance like this, the growth of modern physics would have been 

The Process of Communication 277 

severely hampered when the need to make such a statement did 

In every language, factors comparable to these six, though differ¬ 
ing widely in detail, are found. Redundancy is not an imperfection 
in language, but an essential feature, without which language 
would be inoperative. 

19.14 Communication theory has been developed primarily by 
engineers interested in the design of telephone and telegraph cir¬ 
cuits. Their problem is to produce equipment with the maximum 
efficiency in the use of channels. This necessitates some method 
for measuring theoretical channel capacity as well as the amount 
of that capacity used in the transmission of any given code. The 
usefulness for this kind of problem of calculations of the type just 
described is now established beyond question. The method has 
been elaborated to a very high degree and is being extensively used. 

As linguists have become acquainted with the results, two types 
of questions have arisen: What are the implications of this body 
of knowledge for linguistic theory? Do these methods have any¬ 
thing to contribute to practical problems of linguistic analysis? 

As regards the first question, there seems little doubt that the 
implications are considerable. They are, however, as yet only 
vaguely perceived. This is partly because few linguists have as yet 
more than a very superficial knowledge of communication theory. 
It is also partly because the two approaches are so fundamentally 
different that it will take considerable labor to establish in detail 
the relationships between the two bodies of theory. But in spite of 
this lack of definite evidence, the indications are that communica¬ 
tion theory will make important contributions to linguistic theory. 

Linguistic theory, particularly as developed by American lin¬ 
guists, has been intimately based on analytic method. The first 
question is, therefore, not really separable from the second. The 
crucial question is whether the methods of information theory can 
be brought to bear on specific linguistic problems. That is, are 
there questions about language structure which can be answered 
by the methods of communication theory, and if so, can they be 
handled as well or better than by traditional methods of descrip¬ 
tive linguistics? Work in this field is just getting under way, but 
there is some possibility that linguists will find methods of applying 
the new techniques to their own characteristic problems in a fruit- 

278 The Process of Communication 

ful way. If this occurs, the implications for linguistic theory will 
follow inevitably. 

19.15 In 9.8 a brief mention was made of the Turkish system 
of vowel harmony, which results in a special type of restriction on 
the freedom of occurrence of vowels in successive syllables. For 
example, a number of suffixes have allomorphs containing the 
vowel /ii/ which occur if and only if the preceding syllable con¬ 
tains either /ii/ or /6/. The effect is to produce a very striking kind 
of redundancy. It would seem that this might be a fruitful prob¬ 
lem on which to test the usefulness of techniques based on com¬ 
munications theory in the analysis of a structural phenomenon. 

As an exploratory experiment a sample of 1,000 five-syllable 
words was taken and the vowel sequences in each tabulated. The 
sequences /e i e i i/ and /e i e i e/ were each found to occur 
28 times. 307 other sequences were also found, nearly half of them 
only once each. If we assume that this sample is adequate, the 
total information carried by the vowels in such words is about 
7.62 binits. The theoretical average capacity of the vowels in such 
words would be 15.00 binits. (There are eight vowel phonemes in 
Turkish. If these were equally probable, the eight possible signals 
in any given syllable would carry 3.00 binits of information, and 
five syllables would carry 15.00.) The redundancy is therefore just 
under 50 percent. This docs not seem extreme, though we have 
insufficient data to make comparison with other languages, and 
so do not really know whether such a figure is high or low. 

Suppose that vowel harmony were followed rigorously through¬ 
out the word. The maximum total information carried by the 
vowels of five-syllable words would be 7.00 binits. (The first syl¬ 
lable might have any one of eight vowels. Equal probabilities would 
give the maximum capacity of 3.00. Each successive vowel must 
be either the proper one of /i u i u/ or the proper one of /e a/. 
Thus each successive syllable will carry only 1.00 binit maximum.) 
If this calculated capacity is compared with the observed capacity, 
it will be found that the two are rather similar. This suggests one 
of the following: (1) If vowel harmony even approaches the effec¬ 
tiveness on which we based our calculation, redundancy of other 
sources must be either lacking or very small. Or, (2) if other sources 
of redundancy make any appreciable contribution, vowel harmony 
must be that much less effective. The most fruitful use of the whole 

The Process of Communication 279 

investigation would seem to be the possibility that more detailed 
examination will enable us to determine with some precision the 
actual source of the observed redundancy, and so to see the func¬ 
tional significance of vowel harmony within the total system. This 
would seem to be indicative of the general usefulness of the method. 
Probably there arc few if any structural problems which can be 
directly analyzed by this method, but there would seem to be 
many instances where it will reveal the functional significance of 
some structural detail in the whole system of the language. 

19.16 We may proceed to such a detailed analysis by calcu¬ 
lating the redundancy in individual vowel positions, or even in¬ 
dividual dimensions of contrast, or in various sequences of vowels. 
The amount of detail that must be tabulated is quite large, and its 
full presentation here is not feasible. The following samples may 
however indicate something of the possibilities: 

If we examine the vowel of the first syllable only we find the 
following observed frequencies: 

lx/ 129 /u/ 61 /i/ 38 /u/ 96 

/e/ 267 /6/ 46 /a/ 294 /o/ 69 1 = 2.63 

The redundancy occasioned by disparate frequencies of the eight 
vowels in this position is only 12 percent, which would seem rather 
low. But we need not stop with that. We may note that of the 
vowels, 503 are front vowels and 497 are back. The redundancy 
here is very nearly .00. High vowels total 324, and low vowels 676, 
giving a redundancy of 10 percent. Unrounded vowels total 728, 
rounded 272, producing a redundancy of 16 percent. The largest 
part of the total redundancy observed in the vowels of the first 
syllable is therefore due to the less efficient use of the contrast 
rounded : unrounded, and not at all to the contrast front: back, 
which is used with high efficiency. 

Or we may examine the vowels of each successive pair of two 
syllables. For example, taking the vowels of the next to the last 
and the last syllables we find that the total information is 3.42 
binits, as against a maximum capacity of 6.00. The redundancy is 
43 percent. However, 2.31 binits of information is ascribable to 
the fourth syllable (redundancy, 23 percent), leaving only 1.11 
carried by the fifth vowel (redundancy, 63 percent). The greater 
part of the difference in redundancy is the effect of vowel harmony. 

280 The Process of Communication 

Numerous other calculations can be presented which would 
further identify the various factors which contribute to the re¬ 
dundancy of successive Turkish vowels. There is one difficulty 
common to them all: we do not have enough data from other 
languages to provide a basis for interpretation. Nevertheless, the 
calculations do seem to indicate that the method can, with further 
experience, provide a useful technique for interpreting linguistic 

19.17 In 19.13 we listed some of the sources of redundancy in 
a language. It is quite evident from a consideration of these that 
redundancy is an essential feature of a language. That is, if 
redundancy were eliminated, it would be at the cost of eliminating 
all structure, leaving an utterly unworkable code. Redundancy is 
a consequence of structure, and it is perhaps in this regard that 
linguists are most interested in it. There is, however, another rdle, 
quite important in some instances. 

There is always a possibility that something will go wrong in the 
complicated process of communication. Difficulties can arise in the 
process of encoding or decoding. Signals can be lost in the channel, 
or stray signals can get into the channel and be confused with 
elements of the code. The latter is quite familiar in everyday 
spoken language. It is noise, extraneous sounds that are heard with 
the speech and which we cannot wholly separate from the speech. 
The ultimate effect of all the sources of error is the same, and com¬ 
munication theorists have given a technical definition to noise to 
cover them all. (This is merely another case comparable to the 
redefinition of information and redundancy.) Noise is any un¬ 
predictable interference with a communication system. “Un¬ 
predictable” is the essential item in the definition, since if noise 
could be known in advance, it could be eliminated. 

Noise is potentially present in every communication system. 
Naturally, some systems have more noise than others. The amount 
of noise can sometimes be predicted from the characteristics of 
the encoder, the decoder, or the channel, and noise can be meas¬ 
ured in a statistical way. 

19.18 Noise, of course, detracts from the efficiency of a com¬ 
munication system. Any perfect code (one that uses the full 
capacity of the channel) is unusable in the presence of noise, since 
any noise will alter the message. The obvious way to avoid trouble 

The Process of Communication 281 

is to introduce redundancy. We commonly do this by repetition, 
but any other method would have the same effect. An effective 
code must have sufficient redundancy to compensate for any noise 
in the system. Since a language t 3 -pically has an appreciable 
amount of redundancy inherent in its structure, it is always 
possible to use it in the presence of a moderate amount of noise. 

When the amount of noise is greater, more redundancy may be 
needed. This is a common experience. When hearing conditions 
are bad we commonly find it necessary to spell out proper names. 
(Proper names seem to have less redundancy than the average in 
English.) We may use the familiar names of the letters /6y biy 
sly . . ./. But since these names do not have as much distinctive¬ 
ness as might be wished, we may use such locutions as /bfy— ajzin + 
b5Hstin/, or we may use a special set of letter names with higher 
redundancy /Sybil bSykor 6arliy . . ./. These are merely devices 
to increase redundancy when the noise level is high. 

19.19 Though the redundancy in language is of direct benefit 
to the users, it is something of a nuisance to the linguist in his 
analysis and description. A straightforward description of any 
utterance would completely bog down in attempting to state a lot 
of detail which, because of high redundancy, is of no significance. 
The basic technique of the science of linguistics is actually one of 
sorting out redundant and nonredundant features, so that each 
type can be given the treatment which is most efficient descrip¬ 
tively. This sorting is performed on a number of different levels 
and gives rise to the “alio- and -erne" principle which is so 
fundamental in linguistics. 

We may illustrate by considering allophones and phonemes. 
The differences between allophones of any given phoneme are 
features of speech which are redundant, once both the phoneme 
and the environment are known. By first making certain general 
statements about allophones, we may thereafter eliminate this 
redundanc 3 r from further consideration by the device of mention¬ 
ing only the phonemes. 

Phonemes are not redundant at this level. They may, however, 
be so at some higher level, so we may again repeat the process of 
separating redundant from nonredundant features. For example, 
in Turkish vowel harmony, there is a contrast between /i U i u/ 
and /e a/, but none within these groups. (There is a phonemic 

282 The Process of Communication 

contrast because not all vowels arc subject to vowel harmony.) 
If we consider each group of phonemes as morphophonemically 
the same, perhaps setting up some special symbols such as H 
(high vowel) and L (low vowel), we can rewrite such a word as 
/gostermektedir/ ‘for example’ as /gostLrmLktLdHr/. Such a 
technique will eliminate from our transcription all the redundancy 
due to vowel harmony, reducing the average redundancy of vowel 
sequences in such words by a very considerable amount. 

19.20 It is at the level of phonemicizing that the greatest 
amount of redundancy is removed. Complementary distribution 
is only a special case of redundancy. In effect we ask the question: 
In any given environment, which will occur, A or B? If this proves 
to be predictable in every environment, the redundancy of the 
contrast is 100 percent, and we say they are in complementary dis¬ 
tribution. But this process of classing allophoncs into phonemes is 
only one aspect of the process. Long before this, a great deal of 
redundancy has been eliminated. Our first phonetic transcription 
groups an immense variety of sounds into relatively few groups, 
each of which we transcribe by a single symbol. 

19.21 What is the redundancy of speech at this level? Various 
physical considerations suggest that the total capacity of the 
channel used (all frequencies of sound which the human vocal 
apparatus can produce and the ear hear) is of the order of mag¬ 
nitude of 50,000 binits per second. Considering the phonemes as the 
messages to be transmitted, speech uses something of the order 
of 50 binits per second (less if we take account of redundancy in the 
use of the phonemes). That is, the redundancy of speech at this 
level is in the neighborhood of 99.9 percent. 

This would seem highly inefficient, but when we consider the 
noise that must be overcome it is not unreasonable. In the first 
place, the vocal apparatus imposes some severe limitations oh the 
combinations of signals which can be produced. For example, there 
is a characteristic pitch range for every voice. To control the full 
50,000 binits per second, the encoder would have to be able to 
use every distinguishable frequency, whether within his normal 
range or not, and to produce any combination of these pitches, 
whether they bear harmonic relationships to each other or not. In 
the second place, speech must operate under a wide variety of 
conditions. Reverberation, resonances in the environment, and 

The Process of Communication 283 

numerous other factors all contribute to modify speech sounds. 
That is, even in what we call “quiet,” there are considerable 
possibilities of what the engineer calls “noise.” To convey the 
full 50,000 binits per second, all these environmental factors would 
have to be controlled. In the third place, speech must be possible 
for all members of the community. The smaller and differently 
shaped vocal tract of a child cannot produce exactly the same 
sounds as those of an adult. We must learn to switch codes and to 
identify as /i/ or /s/, or whatever, the vastly different sounds pro¬ 
duced by a small child, an adult, a person with laryngitis, or one 
out of breath from exertion. Moreover, we must be able to do this 
in a quiet room, in a noisy room, in the open spaces, or wherever we 
meet others. The miracle is that we can do this at all. Only the very 
high redundancy of speech at the sub-phonemic level makes this 



! 20 

Variation in Speech 

20.1 One of the most evident facts about speech is its variability. 
If a large body of utterances in a given language is examined, it 
will be found that no two of them are identical — certainly not if 
sufficiently refined methods of measurement are used in observing 
them. Before any description of speech can be made it is necessary 
to bring some sort of order into the data. This can be done only by 
concentrating attention on some particular type of variation to 
the exclusion of all others. Since there are many different aspects 
which may be selected for study, there are several different 
methods of approach to a systematic description of language. 
Each of these is supplemental to all the others; no one can be com¬ 
pletely understood without some acquaintance with the others. 

20.2 Faced with a large body of utterances in a given language, 
the descriptive linguist starts by selecting a sample of the data 
for special study. This he does in a characteristic way, which 
necessarily predetermines the type of results he will obtain. He 
may attempt to eliminate certain types of variation of minor in¬ 
terest to him by restricting his attention as far as possible to 
utterances produced by one speaker under a single set of circum¬ 
stances. Within this narrowed corpus he seeks out expressions 
which differ minimally in content and expression. By comparing 
these he determines the minimal differences in the expression which 
arc associated in any constant way with differences in the content. 


Variation in Speech 285 

By this means he discovers two sets of elements out of which the 
expression structure of the language is elaborated, the phonemes 
and the morphemes. When, either through necessity or choice, he 
docs enlarge his data by examining utterances of other speakers or 
other circumstances, he docs so in such a way as to divert his 
attention as little as possible from these minimal contrastive 
elements and their combinations. 

20.3 Another quite valid, but basically very different, approach 
would be to sort out from the total mass all those sets of utterances 
which are alike in content. This will eliminate one of the variables 
with which the descriptive linguist is concerned. Within each set, 
these utterances can then be compared and the variation studied. 
This may be done by seeking correlations with non-linguistic 
factors, commonly the speaker and the circumstances. Obviously, 
the results are predestined to be fundamentally different from 
those which the descriptive linguist will attain, since the variation 
under examination is precisely that which the descriptivist will 
attempt to eliminate. 

For example, suppose we were to sort out a large number of 
English utterances which we identify in some manner as being I'm 
going home. These will vary among themselves in the general level 
of pitch, in the speed of utterance, in the degree of nasality, and in 
a number of other features which the descriptivist would disre¬ 
gard, since he does not find them to be phonemic. They will also 
differ in certain phonemic features. Some samples will include 
/h6wm/, others /h6m/, still others /h£wm/. Both /gdwiq/ and 
/g6win/ and several other variants will be found. Other parts of 
the utterance will show other comparable contrasts. The problem 
is to reduce these variations, both phonemic and sub-phonemic, to 
order. The objective is an empirical description of the range and 
significance of variation in English utterances that are in some 
way equivalent. Beyond that, we would like to make some gen¬ 
eralizations about linguistic variation as a characteristic feature of 
language. Here is the basis for a second type of linguistic science. 
Since most workers have restricted their attention to single aspects 
of the problem, we lack a general term for the discipline as a whole. 

20.4 Variation can best be systematized by correlation. One 
possibility would be to determine, for example, whether there is 
any correlation in our sample of utterances between /gdwin/ 

286 Variation in Speech 

(rather than /gbwiij/) and /h$wm/ (rather than /h6wm/). I do 
not know what the answer would prove to be, nor has this approach 
been attractive to workers in the field. Another possibility would 
be to seek correlations with known facts outside language. 

Experience has shown that there are certain categories of such 
facts which are particularly useful in the study of speech: the so¬ 
cial context of the specific utterance, the social position of the 
speaker, the geographical origin of the speaker, and the age of the 
speaker. Each of these provides a useful body of generalizations. 
Moreover, each of these correlations is of sufficient general interest 
that a body of prescicntific folklore has arisen. In addition, there 
is variation which correlates with the individual identity of the 
speaker, a fact which is also well known and socially significant. 
Specific scientific study has, however, not been generally profitable, 
except in regard to “speech defects” (individual peculiarities of a 
type which are a social handicap). Beyond this there is a residue of 
variation which seems entirely random (not correlated with any 
known factor). This is the fraction of the data which is of no in¬ 
terest in the present problem, as linguistically unconditioned sub- 
phonemic variation is of no interest to the descriptivist. It may, 
of course, be of interest in some other approach; for example, it 
could well be that some of this variation is correlatable with 
atmospheric conditions — pressure, humidity, temperature, etc. 
— in which case it might well be of great interest to an acoustic 

20.6 It is characteristic of descriptive linguistics that the data 
is handled in a specific way. Typically, a given segment either is 
or is not a given phoneme. When, as in the case of English /£/, 
there is a delicate balance of evidence for the interpretation either 
as /£/ or as */W> we do not make a statistical statement. Instead, 
descriptive linguists unequivocally treat it either as a unit pho¬ 
neme or as a cluster. Descriptive linguistics is an either-or proposi¬ 
tion, and its methods are applied only where the data can be so 

By contrast, the study of other types of variation in speech is 
thoroughly statistical in its requirements. Much of the data is 
observed in terms of continuous variation, as when we make 
observations on the degree of aspiration in stops, on the pitch of 
the voice or the duration of an utterance, or the like. This must be 

Variation in Speech 287 

correlated with non-linguistic facts which are also subject to con¬ 
tinuous variation. There are in many cultures no sharp delimita¬ 
tions of social classes, but gradations of classes. There is no 
discrete territorial organization into dialect areas, but a geo¬ 
graphical continuum. The vast majority of the data is subject to 
continuous variation and can be prepared for analysis only by 
statistical procedures. 

The method of correlation which we must employ is statistical 
in nature. Moreover, the statistical methods which are needed are 
often complex. With a large number of factors to be considered, 
all of which operate on the same material, speech, it is not sur¬ 
prising that the isolation of individual factors and the assessment 
of their effects is no simple procedure. 

It is this difference in basic method, more than anything else, 
which makes publications from this division of linguistics seem 
so utterly different from those in descriptive linguistics. The differ¬ 
ence is not superficial or incidental; it is the result of fundamental 
differences in the data under examination. 

20.6 Basic to all the problems of language variation is the com¬ 
plex process of linguistic change. This is no single entity ; but the 
cumulative effect of a number of quite different processes operating 
more or less independently. We will discuss only four of them: pho¬ 
netic change, phonemic change, analogic change, and borrowing. 

20.7 Phonetic change may perhaps best be described in the 
form of a specific and more or less artificial example. Consider a 
language in which there are two phonemes: /t/ which is fortis, 
and /d/ which is lenis. A lenis stop is one in which the contact of 
the lower articulator (here the tongue) against the upper articu¬ 
lator is relatively weak. If the strength of articulation of a long 
series of apical stops is measured, we may expect to obtain a fre¬ 
quency distribution something like the following: 

N Idl 

(It is not unusual for the distribution curves to overlap. Re¬ 
dundancy, operating at various levels, will preserve the work- 

288 Variation in Speech 

ability of the system, provided the overlap does not get too 

A given speaker maintains this distribution by monitoring his 
own speech.* That is, he compares the sounds he hears himself pro¬ 
duce against a set of norms. This is actually his own crude in¬ 
tuitive statistical summary of his past experiences with the sounds 
of the language. Now suppose some unknown factor (or at least 
one outside our range of study) causes some of the speakers in 
this community to average slightly more lenis in their pronuncia¬ 
tion of /d/. These more lenis pronunciations will figure in the 
total experience from which the norm is determined. The normal 
range of variation of /d/ will shift slightly in the direction of more 
lenis pronunciation. If this continues, /d/ may become so lenis 
that it is frequently incompletely closed, that is, a fricative [<5] 
instead of a stop [d]. The shift may continue until [5] is much 
the commoner pronunciation. 

What we have just described is a phonetic, not a phonemic, 
change, /d/ still continues to contrast with /t/, just as before. 
Phonetically it has shifted from a stop to a fricative, but unless 
something else has also happened to change the patterns of con¬ 
trast, there is no change in the phonemic status. 

20.8 There are two important characteristics of such a change 
that require comment. In the first place, what is shifting is not 
the pronunciation of a specific sound in a specific place, say a 
certain word. If it were, we might expect the same sound to change 
in a different way in some other place. Instead, the shift affects 
the statistical norm based on all occurrences of the given phoneme 
in a given environment — that is, on all occurrences of a certain 
allophone. In turn this norm controls the pronunciation of this 
allophone whenever it occurs. Phonetic change, therefore, affects 
allophones as wholes. Within the understanding that the effect 
is statistical, phonetic change affects any given allophone con¬ 
sistently. This is commonly expressed by saying that phonetic 
change is regular. This means that any phonetic change will 
affect all instances of the sound concerned in the positions in which 
it is operative. The same phonetic change may affect all the 
allophones of a given phoneme, or only a single allophone. 

The second significant characteristic is that phonetic change 
is a social phenomenon. The statistical norm which controls the 

Variation in Speech 289 

pronunciation of each allophone is not based on one person’s 
pronunciation, but to some extent on the speech of every indi¬ 
vidual whom the speaker hears. Not all of them will have an equal 
effect on the norm. In general, speakers of higher prestige will exert 
more influence than those of lower. The frequency of contact will 
also be very important. Thus, phonetic changes within the speech 
of any intimate group are very likely to be shared. When contact 
is less intimate, greater differences in the rate or even the direction 
of change can be expected. 

20.9 To continue the example of 20.7: Suppose that the shift 
from stop to fricative pronunciation of /d/ affected only /d/ be¬ 
tween vowels. At the end of the process described, we will have 
three phonetic ranges: 

Itl [d| [5] 

There are, however, still only two phonemes, since [d] and [<5] are 
aliophones in complementary distribution. Now suppose that 
through some other process, which has been going on concur¬ 
rently, some of the vowels disappear. The result might be: 

/VdVCV/ [V3VCV] becomes [V3CV] 

/VdCV/ [VdCV] becomes [VdCV] 

[d] and [<5] are now in contrast, hence separate phonemes /d/ 
and /*>/. Phonetic change has led to phonemic change. In this 
instance, as is the most common case, phonemic change is not pro¬ 
duced by a phonetic change affecting the changing phoneme, but 
by the change of some factor which conditions aliophones. With 
this change, new patterns of contrast arise, and the aliophones 
become phonemes. 

We may give an example from English. At one time English had 
a phoneme /n/ with aliophones [ij] before /k/ or /g/ and [n] 
elsewhere. Thus sing was something like /sing/ [sirjg]. Then under 
certain conditions final /g/ dropped, [sfijg] thus changed to [siq], 
and so came to contrast with sin [sin]. This raised [ij] to the status 

290 Variation in Speech 

of a phoneme /q/, but also left /rj/ with a rather peculiar dis¬ 
tribution; mainly it occurs in those places where /g/ dropped, or 
where it is followed by /k/, or in a few places where it has appeared 
by other changes. 

Since phonemic change is generally the cumulative result of a 
series of phonetic changes, it is also regular. That is, whenever the 
proper conditions obtain, phonemic change occurs without ex¬ 

20.10 Analogic change is very different in its mechanism and 
effect. The process is a familiar one. Let us take an example. Sup¬ 
pose that a speaker does not recall the plural of mouse, and fol¬ 
lowing the prevailing pattern in English says /m&wsiz/. The 
chances are that this will not be copied, and hence will not con¬ 
tribute to a permanent change in the language. But if various 
factors combine in the right way, the outcome might be different 
and /m&wsiz/ might establish itself as the regular plural of mouse. 

The noticeable fact about this kind of change is that each 
individual change stands or falls on its own merits. There is no 
reason to expect that, if this change should establish itself, louse 
would follow the same analogy. The plural might continue to be 
/l&ys/, or change to /l&wziz/ on the analogy of houses, or some 
more obscure analogy might become the pattern. Analogic 
change is not regular. It may, and frequently does, produce 
an increase in irregularity, but it may equally exert a regularizing 

20.11 Borrowing is just what its name implies — the copying 
of a linguistic item from speakers of another speech form. The most 
evident instances are those in which the two forms of speech are 
quite different. The loan words sometimes preserve characteristics 
by which their foreign origin can be readily discerned, as in the 
case of hwana, a recent introduction to American English whose 
Swahili origin is marked by the peculiar initial cluster /bw/. In 
other cases loan words are made to conform more closely to the 
phonologic or morphologic patterns of the language, but such cases 
can usually be easily identified by tracing their etymology. The 
commonest case, however, is that of loan words in one dialect taken 
from another closely related dialect. These instances may be very 
difficult to identify, often appearing as minor exceptions to other¬ 
wise regular phonemic changes. Just as with analogic changes, 

Variation in Speech 291 

borrowing is a more or less random and unsystematic process. 
Individual items are involved, seldom definable groups of words. 

20.12 With these four types of linguistic change in mind, we 
can now examine briefly some of the features of linguistic varia¬ 
tion. The most spectacular and best known is that between geo¬ 
graphical forms of speech. When the differences are small, these 
are known as dialects. When larger, they are known as languages. 
However, no exact definition of these two terms is feasible. They 
have for so long been used in widely varying meanings that it is 
nearly hopeless to impose any uniformity on their usage, even if 
some suitable criterion could be found. 

The process of dialect formation can best be understood if we 
hypothecate a socially homogeneous population with only local 
mobility and speaking at the start a more or less uniform language. 
(Needless to say, such a situation is a highly hypothetical one!) 
Now consider what will happen if some conspicuous phonetic 
change gets underway at some one spot in the territory. As we 
have noticed, phonetic change is a social affair. It will be shared 
with any speaker with whom intimate contacts are maintained, 
and there is considerable probability of its being shared to some 
extent by speakers with whom contacts are more casual. Either 
the innovation will spread into areas surrounding the center of 
origin, or the influence of the speech of the surrounding areas will 
lead to its suppression. If it is successful in spreading, we should 
expect the change to have proceeded farthest in the center of 
origin. Away from the center the change should become less and 
less conspicuous. There would, however, be a gradual lessening of 
its influence, rather than any abrupt ending. We can make an 
analogy as follows: Consider a flexible membrane, perhaps of 
rubber, stretched across a frame. Then consider a body pushing 
against the memhrane at some point. The membrane is not torn, 
but it is displaced from its original position by an amount that 
decreases as you move away from the center of disturbance. In the 
figure on page 292 we give diagrams indicating the situations at 
two successive stages of this process. We can map the effect by 
noting the points at which a certain degree of change has been 
attained. A line indicating the limit of some stated degree of 
linguistic change is known as an isogloss. As the process of change 
continues, the isogloss would seem to move outward, away from 

292 Variation in Speech 

the center of innovation. On the diagram, we have indicated the 
position of the isogloss at each of the two stages, together with 
its apparent movement. 

20.13 A linguistic change in a more or less homogeneous popu¬ 
lation can be most easily described in terms of the appearance and 
movement of isoglosses across a map. These isoglosses are, of 
course, statistical abstractions which cannot be directly observed. 
The precision of our analogy, or of the diagrams accompanying it, 
is never found. When such a line is drawn, it merely indicates the 
compiler’s conclusion from the data at hand that within the line 
one pronunciation is commoner, and outside, another. As a change 

Variation in Speech 293 

spreads, there are always some speakers who maintain the old 
form longer than their neighbors. There may be individuals out- 
side the area which we limit by an isogloss who have adopted the 
new feature ahead of their neighbors. 

An isogloss is a representation of statistical probabilities. As 
such it is a convenient means of description, but may be misleading 
if the apparent sharpness of distinction between the areas is not 
carefully discounted. The drawing of isoglosses is one of many 
places where it is easy to be over-precise. The reading of them is 
even more dangerous, since the reader has not seen the intricate 
mass of data upon which they are based. 

20.14 We can add one bit of realism to the picture just drawn 
by suggesting that the density of communication is not everywhere 
the same. Suppose for example that the language area is bisected 
by a barrier of some sort — perhaps a large river, a mountain 
range, or a major political boundary. As a result, there is less com¬ 
munication between people across the barrier than on either side. 
Barriers are seldom absolute, however, so that there is typically 
some communication. Now consider some innovation spreading 
from person to person as they communicate with each other. The 
effect may be pictured in terms of an isogloss moving across the 
territory. When this isogloss reaches the region of low density of 
communication (the barrier), its progress will be retarded. We 
may expect it to take considerably longer for the isogloss to cross 
the barrier than to traverse an equal distance elsewhere. 

If instead of one isogloss there are a number of them moving 
across the area at the same time, we may expect that they will all 
tend to be retarded at the barrier and hence pile up. The result is 
what is sometimes referred to as a bundle or fascicle of isoglosses. 
The history of the individual isoglosses in such a bundle may be 
quite different. Some may have been arrested in approximately 
their present position for a long time; others may have just arrived. 
Some may even be in process of comparatively rapid transition 
across the barrier. Some may be moving in one direction, and some 
in the other. Often it is impossible to trace all the factors which 
account for the particular assortment of isoglosses observed at one 
place. Nevertheless, any geographical area of low density of com¬ 
munication is likely to be marked by such a bundle of isoglosses. 

20.15 Since the presence of an isogloss is a graphic way of por- 

294 Variation in Speech 

traying a transition in speech characteristics from one area to an¬ 
other, a bundle of isoglosses may be interpreted as marking a zone 
of relatively great transition in speech. We may, therefore, think 
of it as indicating a dialect boundary. Such a boundary is rarely, 
if ever, sharply abrupt. In the first place, the individual isoglosses 
represent transitions that are seldom other than gradual; com¬ 
monly there are quite clear exceptions on either or both sides of 
the line. In the second place, the various isoglosses in any bundle 
seldom coincide exactly. As you pass from one dialect area to an¬ 
other, you will commonly observe the appearance of one new 
feature after another. There is almost never any one place at which 
one dialect gives way suddenly to another, and which would show 
on a map as the coincidence of a number of isoglosses of major 

Not infrequently a bundle of isoglosses will run along rather 
close together for a distance and then fan out. The classic case is 
that of a bundle of isoglosses running across Germany separating 
the “northern” dialects from the “High German" dialects. For 
most of the distance the boundary is quite sharp. That is, the 
isoglosses by which the dialectal differences can be described run 
rather close together, though they do cross and recross each other 
in an intricate tangle. But as they approach the Rhine Valley, the 
various lines separate. Some continue about the same trend, others 
turn southward, and still others northward. That is to say, in the 
Rhine Valley the boundary is much less sharp than it is to the East. 

20.16 A typical instance of a dialect boundary of the sort that 
we might expect from the description above runs across Virginia 
along the Blue Ridge Mountains. From the Potomac to the 
Roanoke, the Blue Ridge is a relatively narrow and sharply de¬ 
fined range rising up suddenly from the Piedmont on one side and 
the Shenandoah Valley on -the other. Crossings are relatively few, 
and none are particularly easy. The higher ridges are unoccupied, 
and are now largely national park. Particularly prior to the de¬ 
velopment of modem transportation, the two halves of the state 
lived rather separately from one another. It is, therefore, not sur¬ 
prising that a major dialect boundary should lie here, separating 
the Midland group of dialects from the Southern group. 

However, the physical boundary is not the whole basis for dis¬ 
tinction, as there are other physical boundaries which are just as 

Variation in Speech 295 

effective in restricting communication but are of less linguistic 
significance. The Appalachian Ridges to the west of the Shenan¬ 
doah Valley are a case in point. Moreover, some major dialect 
boundaries have no obvious physical basis. An example of the 
latter type is the boundary running east and west across the middle 
of Ohio, Indiana, and Illinois, and which may be conveniently 

Major Dialect Areas in the Eastern United States 
(The westward extent of these areas is not yet known.) 

described as approximately following U.S. Highway 40. While not 
as sharp as the boundary along the Blue Ridge, it is quite evident, 
and the speech difference between the northern and southern parts 
of these states is quite marked. The basis for this boundary is of 
quite a different sort, since there is no important physical barrier. 

There are three major dialect areas in the eastern United States. 
They developed out of the three major centers of settlement on the 
Eastern seaboard. The westward expansion carried New Eng- 

296 Variation in Speech 

landers across the northern Middle West. Pennsylvanians crossed 
the mountains into the southern Middle West and followed the 
valleys southwestward into the Appalachian Highlands. The 
Southern Coastal settlements spread inland to the mountains and 
southward along the Piedmont into the Gulf States. The result is 
the three major dialect areas approximately delineated on the 
map on page 295. While this is greatly oversimplified (for exam¬ 
ple, Southern elements are quite as prominent as Pennsylvanian 
in much of the Midland), it does serve to indicate that historical 
factors are often of major importance. In the United States, one of 
the chief means of movement of isoglosses has been the movement 
of the people themselves. 

20.17 Because of the complexities of topography, settlement 
history, interregional communication, and prestige of regional 
centers, dialect boundaries are often vague, complex, and difficult 
to delimit. Dialects are seldom subject to the neat classification 
which we might desire, and which the layman sometimes imposes 
upon them. For example, American folk-linguistics recognizes two 
major dialect areas, "Southern” and "Northern.” But there is no 
discernible linguistic division at or near the Mason-Dixon line. 
"Southern” dialects are exceedingly diverse. The sharpest dialect 
boundary in the United States runs directly through the South 
roughly along the Blue Ridge mountains. A "Northern dialect” 
is as much a fiction as a "Southern dialect.” There are greater 
similarities of speech between southern Indiana and the Blue Ridge 
of Virginia than between the former and northern Indiana or the 
latter and eastern Virginia. 

It would be nearly as fallacious to speak of three dialects, 
Northern, Midland, and Southern, unless it is understood that each 
of these is more properly a group of dialects, and that the bound¬ 
aries between the major areas vary from rather sharp in a few 
localities (such as the Virginia Blue Ridge) to very indistinct and 

20.18 If we examine the difference between any two dialects, 
preferably separated by quite a distance, we find that they are of 
two kinds. Some are quite regular in the sense that they are re¬ 
peated over and over and may be observed almost universally in 
all words of a certain kind. These are largely phonetic and pho¬ 
nemic differences. For example, those words pronounced with /aw/ 

Variation in Speech 297 

in New England and upper Middle Western dialects are mostly 
pronounced with /®w/ in certain dialects of the coastal South. 
The same items are pronounced either with /ow/ or /aw/ in 
eastern Canada. The latter variation is not capricious; /ow/ occurs 
generally before voiceless sounds, /aw/ before voiced. 

An example of a phonetic difference is the pronunciation of /$/ 
used by some speakers in the Metropolitan New York area. This 
is a dental stop, rather than a dental fricative. Because it is dental 
rather than alveolar it contrasts with /d/, so there is no phonemic 
difference at this point between this form of speech and that found 
elsewhere. Non-New Yorkers, however, commonly misinterpret 
such a /$/ as a /d/, since the phonetic basis of the contrast is quite 
different in their speech. 

20.19 Since analogic changes are independent of each other, 
dialectal differences of this origin cannot be expected to show any 
great regularity. For example, the past tense holp /h6wp/ for the 
verb help is now widespread only in the South and the southern 
Midland. It represents the older form; helped /h61pt/ has replaced 
it by analogic change in most other dialects. By contrast, the 
analogic formation knowed /n6wd/ is much more common in the 
South, though found sporadically throughout the East. Another 
new analogic formation, drinked, is found rather commonly in 
northern New England, in the Chesapeake Bay region, in the 
Carolinas, and occasionally through the remainder of the East. 

It is, thus, not passible to make such generalizations as can be 
made with phonetic and phonemic differences. A familiar notion 
of American folk-linguistics is that the Southern highlanders speak 
“pure Elizabethan English.” This is manifestly untrue. They do 
have a number of conservative traits of language, as for example 
holp (which they share with many lowland Southerners), but these 
are counterbalanced by many innovations not found outside the 
area, as well as by innovations that they share with other regions. 
“Pure Elizabethan English” became extinct with the Eliza¬ 
bethans, but its elements persist, mixed with various innovations, 
in ai.l English dialects. There is no reason to expect a significantly 
different proportion of such survivals in the Southern highlands 
than would be found elsewhere, but the assortment of features 
which have persisted may be expected to be rather different from 
one area to the next. 

298 Variation in Speech 

20.20 Another complication in any neat picture of geographic 
dialects arises from the fact that the people in one community do 
not all speak alike. In a culture such as ours, in which there is a 
measure of class distinction, the most obvious difference is likely 
to be that between classes. Not infrequently there is more differ¬ 
ence between upper- or middle-class speakers in one community 
and the lower-class speakers of the same community, than there is 
between the upper- or middle-class speakers of two rather distant 
communities. For example, holy /h6wp/, though quite widespread 
throughout the South, is seldom used by the educated, urbanized 
middle class. They say helped /h&pt/, much as would a New 
Englander of the same class, and they tend to look on /h6wp/ as 
rustic or ungrammatical. It is therefore not possible to speak of 
/h6wp/ as a Southern expression without some qualification. 

In any given area, it is possible to recognize and describe a sys¬ 
tem of social dialects. These cannot be expected to be clearly dis¬ 
tinct from one another, but the concept is in general a useful one. 
As a general scheme, three such are commonly listed: cultivated 
speech, common speech, and folk speech. Cultivated speech is 
characteristic of educated, urbanized, and generally middle- or 
upper-class people. It is that which is taught, with various degrees 
of success, in the schools; it is used on the radio and generally 
carries high prestige. Folk speech is characteristic of isolated rural 
people and is fast disappearing in many areas. Common speech is 
merely a convenient label for the greater part of the range of inter- 
gradation between the other two. 

In general, geographical dialectal differences arc most pro¬ 
nounced in folk speech, and least pronounced in cultivated speech. 
Nevertheless, there are regional variations even in the latter. In 
the Southern mountains, certain Midland locutions which are not 
greatly different from cultivated speech farther north are con¬ 
demned because they do not conform to a cultivated speech 
standard largely based on lowland Southern patterns. Certain 
characteristics of this Southern cultivated speech would be simi¬ 
larly condemned in other regions. While cultivated speech is more 
uniform than folk speech, the standards that prevail in various 
parts of the United States are different and occasionally con¬ 

20.21 Possibly as a result of differences of social dialect, there 

Variation in Speech 299 

are also differences in levels of speech. By this we mean that a 
single speaker, without departing from the conventions usual in 
his area, speaks differently in different social situations. Different 
vocabulary, constructions, and even pronunciations are used in 
formal address and in familiar conversation. Such differences are 
found in the speech of all Americans. Probably they are somewhat 
less conspicuous with speakers of folk speech than with those who 
use cultivated speech (with the exception of a few of the latter who 
use a formal style in all circumstances and thereby appear affected). 

In general, formal speech tends to approach cultivated speech 
more closely; informal speech tends to conform more to folk usages. 
Particularly in the case of educated and urbanized speakers, the 
most conspicuous difference may be greater conformity to written 
usage in the case of formal style. 

20.22 No discussion of dialectal differences, however brief, 
should omit some mention of eye dialects and special dramatic 
dialects. By the former we designate certain quite conventional 
spellings which are used in written literature to indicate that the 
speaker is using folk speech. These include such items as hoss for 
horse, massa for master, sez for says, wuz for was, etc. Some of these 
represent, not always very well, actual differences in pronunciation 
between folk speech and cultivated speech, or between different 
geographical dialects. A surprising number of them, including sez 
and wuz, have no such basis at all. /sez/ and /woz/ are the com¬ 
mon pronunciations of both words in all American speech, folk 
and cultivated. The use of these spellings to indicate folk speech 
is therefore purely conventional. Eye dialect is not, therefore, to 
be considered as an actual portrayal of folk or regional speech so 
much as a stylized literary device to signal that folk speech is in¬ 
tended. It may also be interpreted as evidence on the linguistic 
folklore of Americans, and it indicates how unrealistic, and often 
prejudiced, our linguistic notions generally are. 

Much the same can be said about dramatic renditions of dialects. 
While some actors are excellent mimics and know some dialects 
thoroughly, most use an artificial and conventional stage dialect; 
no Irishman speaks like the typical stage or radio “Irishman.” 
Increased public sensitivity has cut down some of the crudities of 
representation of certain ethnic and national minorities. But the 
characterizations of “country rubes” are in general a most marvel- 

300 Variation in Speech 

ous garbling of discordant elements from various folk dialects and 
from no identifiable source. 

Both these phenomena arc evidence of a tremendous popular 
interest in speech variation. Probably no other aspect of linguistics 
has so great a popular appeal in this country. It is unfortunate that 
the general public should be misinformed on most aspects. What is 
needed is a more intelligent and appreciative attitude toward 
dialect, speech levels, and individual speech characteristics. 




Writing Systems 

21.1 Written communication must be sharply distinguished from 
spoken. The common tendency to use “language” to refer to either 
indiscriminately has so frequently given rise to serious confusion, 
not merely among lay people, but also among professional linguists, 
that many are reluctant to use it of any written code at all, even 
with explicit qualification. Many linguists consider all forms of 
writing entirely outside the domain of linguistics and would re¬ 
strict the discipline to the consideration of spoken language only. 

Nevertheless, the relationships between speech and writing are 
close and intimate. Many of the same methods of study can be 
used in dealing with both, and the structures revealed are in many 
respects similar. But if both are to be treated within the framework 
of a single discipline, it is essential that they be clearly distin¬ 
guished at all times. The term language, when used in any linguistic 
context without qualification, should be reserved exclusively for 
vocal language, that is, for communication by means of speech. The 
qualified term, written language, will be used here, in default of 
any other unambiguous term, for a total system of communication 
based on writing. 

21.2 A written language includes as one level of its structure a 
writing system. This term will refer to a system of conventions in 
the use of certain symbols as the basic signals in a code known as a 
written language. For example, the conventions of English spelling, 


302 Writing Systems 

conceived as patterns of uses of the familiar letters of the alpha¬ 
bet, are a part of the English writing system. In the same way, the 
somewhat different conventions of Dutch spelling constitute an¬ 
other writing system, though the alphabet used in both is the 
same. The difference in writing systems is apparent in the evident 
non-English appearance of such words as zijn, hoar, nieuwe, etc. 
They contain sequences of letters which arc unusual or quite 
strange to the English writing system. 

21.3 Not only docs a writing system have its own structure 
which can be studied, but there is also a set of conventions of rela¬ 
tionship between the writing system and certain structures (com¬ 
monly phonologic) in an associated spoken language. These also 
require discussion, but it is essential to avoid confusing these rela¬ 
tionships with the conventions internal to either code. We refer 
to these relationships between structures in written languages and 
structures in spoken languages as fit. This chapter will be con¬ 
cerned with writing systems and the fit between them and the 
spoken languages. 

21.4 A writing system consists of a set of graphemes plus cer¬ 
tain characteristic features of their use. Each grapheme may have 
one or more allographs. The graphemes and allographs have a 
place in the writing system comparable to that of the phonemes 
and allophones in the phonology, and the relationship of graphemes 
to allographs is comparable to that between phonemes and allo¬ 

A convenient example may be drawn from the familiar form of 
written Greek. As used in modern printed books there are two 
forms of the letter sigma. At the ends of words it is written s else¬ 
where <r. These two symbols are in complementary distribution 
and have a similar reference to the phonology of spoken Greek. 
They are therefore considered as allographs of a single grapheme, 
(<r). (Graphemes may be indicated by a notation included in ( ). 
This is comparable to the notation / / and { } for other funda¬ 
mental structural units.) A series of minimal and subminimal pairs 
will easily establish the graphemic contrast between cr and most 
other symbols in the system. The process is basically the same as 
that by which the phonemes of a spoken language are established, 
and most of the same problems arise. 

For example, only svpxxii'l'MVMOV occur at the ends of 

Writing Systems 303 

words. (That is, before space; in a discussion of writing systems, 
word has the quite specific meaning of a portion of text marked off 
by space or comparable markers, and does not give rise to the 
difficulties which are seen in spoken languages.) s is therefore in 
complementary distribution not only with <t, but equally with ten 
other symbols, ^yd^OXfxTrrtp. An analyst must, therefore, make 
the decision with which of these to class s. This is obvious to any¬ 
one who knows the Greek alphabet, but largely because his first 
introduction to a and s was as alternative forms of a single letter 
with a single name, sigma. An independent decision would have to 
be based on the fit of written to spoken Greek, internal structure 
of the graphemic system, and changes of graphemes in combina¬ 
tion — in much the same way that comparable phonemic problems 
might be solved in the light of phonetics, phonemic structuring, 
and morphophonemics. 

The Greek writing system also makes use of the symbol 2. This 
shows special relationships to <r, exactly comparable to those 
existing between A : a, B It will be found best to treat 

2 as consisting of two graphemes (a) and {-} (capitalization) 
and all the others in the same way. This is analogous to the way 
to which pitch and vowel quality are commonly separated in a 
phonemic analysis of a spoken language. That is, capitalization is 
in many ways comparable to a “suprasegmental” phoneme. 

21.6 Typically (and as a first approximation for our discus¬ 
sion), each grapheme represents some portion of the structure of 
the associated or underlying spoken language. The latter is the 
reference of the grapheme. Since the spoken language expression 
consists of two major structural systems, there arc two major types 
of graphemes from the point of view of their references. 

The most familiar type of grapheme is that with a phonemic 
reference. The three letters in box represent English graphemes of 
various subtypes of this type. The reference of 6 here and of (b) 
generally is /b/. There is some, but relatively little, variation. The 
reference of x here is /ks/, a sequence of phonemes. A phonemic 
reference need not be a single phoneme, but can be any phono- 
logically definable structure. The reference of o in this instance is 
/a/ (in some dialects /o/), but the grapheme (o) has a very con¬ 
siderable number of references, including commonly /ow/, /a/, 
/a/, /o/, etc. The reference of a grapheme may be single-valued 

304 Writing Systems 

or multi-valued. These complexities are merely instances of the 
intricate lit which exists between the English writing system and 
English phonology. Yet in spite of the complexities, the references 
of most of the graphemes are basically phonemic. In Swahili the 
spelling ng refers to /qg/, whereas ng' refers to /q/. <’) has a 
phonemic reference, though of a peculiar sort, since it marks the 
absence, not the presence, of a phoneme /g/. 

21.6 A second type of grapheme has a morphemic reference. 
This is the case with English &. The reference is typically to the 
morpheme {and}; not, it is important to note, a sequence of 
phonemes /lend/. Two lines of evidence present themselves. First, 
& may be read /ifcnd/ /on/, /iq/, or as any other of the numerous 
allomorphs of {and}, and in no other way (if we overlook the in¬ 
frequent case where it is read /et/ — another morpheme — in the 
special context &c.) Second, if & represented a sequence of pho¬ 
nemes, then spellings such as *s& for sand, *h& for hand and 
*&rcw for Andrew would be possible. But no such spellings occur. 
(Note: The argument is not that *s& would occur, but that 
spellings of this sort could occur. English spelling is sufficiently 
arbitrary that we cannot base anything on the non-occurrence of 
any specific possible spelling.) 

Another somewhat different instance of an English grapheme 
with morphemic reference is English ' in boys’. The apostrophe 
here indicates the presence of a morpheme which consists of a 
phonemic 0. In boy’s, ’ also has a morphemic reference since it 
serves to mark the morpheme {-£*}, here spelled as distinct 
from {-Zi}, which would be spelled -8 in this context. Boys, boy’s, 
and boys’ are phonemically identical, but morphemically distinct. 
Therefore, any graphic difference must be considered as having a 
morphemic reference. 

21.7 Graphemes with morphemic reference are commonly 
called ideograms. These are defined as representing an idea, and 
this definition is often interpreted as implying that they have no 
direct connection with spoken expression. A possible more precise 
use of the term ideogram would be to label graphemes with mark¬ 
edly multi-valued morphemic reference in distinction from those 
with more nearly single-valued reference. 4- may be taken as an 
example approximating this in English, since it can be read plus, 
and, more, etc. All these are phonemically and morphemically quite 

Writing Systems 305 

different, but have some contact on the content plane. The refer¬ 
ence of 4- might be taken to be the common content in plus, and, 
more, etc. But to do so is probably unnecessary. Instead, + differs 
from & only in having a greater variety of morphemic references; 
that is, + has a multi-valued morphemic reference; that of & is 
nearly single-valued. 

21.8 Chinese is the typical example of a writing system which 
is alleged to be ideographic. If this implies any lack of precision 
in the reference of the characters, it is not ideographic. The ma¬ 
jority of the graphemes (of which there are necessarily a very 
large number) have unambiguous morphemic references. In any 
given dialect of Chinese, any given symbol will be consistently read 
in the same way — that is, morphemically the same: there may 
be allomorphic variations. 

The Chinese writing system is generally assumed to have de¬ 
veloped from pictorial representations. Some of the signs probably 
went through a stage of rather loosely defined reference to content 
rather than expression before each became tied to some specific 
morpheme. Nevertheless, the Chinese writing system seems to 
have had basically morphemic reference since its beginning as a 
system. (A rebus is not a writing system, but a puzzle. As long as 
the Chinese characters were merely rather vaguely defined pic¬ 
torial representations, they were not a writing system, but some¬ 
thing more nearly comparable to a rebus.) In expanding the stock 
of characters to the present stage where there is a grapheme for 
almost every morpheme, recourse was had to ideographic and 
phonologic devices. Popular discussions of Chinese writing systems 
for Westerners commonly dwell upon these. They do make an inter¬ 
esting presentation. But the ultimate result was a system of 
graphemes, each of which (with unimportant exceptions) has a 
specific morpheme as its reference. That some of the graphemes are 
compound in origin is evident, but more of historic interest than 
of structural significance. (For that matter, w, called /d^bilyffw/, 
is obviously compound in origin, but that does not have appre¬ 
ciable effect on its structural significance in the English writing 

21.9 The peculiar nature of the Chinese writing system has so 
affected thinking about the language as to deserve a somewhat 
parenthetical digression. Many Chinese morphemes consist of a 

306 Writing Systems 

single syllable. Since they are represented in writing by a single 
grapheme, it early came to be the convention to consider each 
character as representing a syllable, and vice versa. As a result, 
two-syllable morphemes were written by two-character graphemes, 
and these were in turn interpreted as representing two separate 
vocabulary items. Since, moreover, each character in a text is 
written separate from all others, and no groups of closely associated 
characters are set off by spaces from other such groups, it became 
traditional to equate the characters with those portions of the 
text which are written separately in European languages. As a 
result each Chinese character is commonly said to represent a 
“word.” Hence there has arisen the legend that Chinese is “mono¬ 
syllabic,” that is, that each word consists of a single syllable. 
Nothing is farther from the truth. There are numerous instances of 
so-called “words” which never occur except with some other spe¬ 
cific “word.” For example Ht occurs only with Jifl, and vice 
versa. The combination means ‘coral’ and is read shanhu. 
If neither occurs without the other, there can be no basis for 
separating them in any analysis of either the spoken or the writ¬ 
ten language, except of course on the a priori assumption: char¬ 
acter - syllable ■ word. The two individual syllables are neither 
words nor morphemes, and the two characters are not separately 
graphemes. Instead there is one grapheme, traditionally written as 
two characters, and its reference is one two-syllable morpheme. 
The confusion, and this is merely the beginning of it, arises largely 
from a failure to keep spoken and written Chinese clearly apart as 
distinct, though related, codes. 

21.10 Writing systems such as that of Chinese, where the 
majority of the graphemes have morphemic references, are com¬ 
paratively few in number. More often the graphemes typically 
have references within the phonologic system of the spoken lan¬ 
guage. Individual graphemes can stand for individual phonemes, 
or for sequences of phonemes. When the former is the case pre¬ 
dominantly, the system is called alphabetic writing. In the latter 
case the references are usually definable types of sequences which 
may be called syllables, and the system is a type of syllabic writing. 
The total list of such graphemes is called a syllabary, this term 
taking the place of the familiar “alphabet.” 

In speaking of syllabic writing systems, the term syllable is used 

Writing Systems 307 

in a specific technical sense as that portion of the stream of pho¬ 
nemes in speech to which a grapheme has reference. It is usually 
definable for any given language and writing system. Most com¬ 
monly each syllable consists of one vowel and all preceding con¬ 
sonants. The division into syllables for the purpose of the writing 
system may be largely independent of any division which might be 
suggested by any direct examination of the spoken language. That 
is, from the point of view of the phonology, it may be wholly or 
largely an arbitrary division. 

21.11 One of the best examples of a syllabic writing system is 
that used for some time in writing the Cherokee language. It was 
invented in 1821 by Sequoya, for whom quite deservedly the giant 
trees of the Pacific coast were named. It proved to be of great use¬ 
fulness to the Cherokee people for many years. The table gives 
the signs used together with a conventional romanization. The 


























































































































(T 3 
























3 ka 
tr hna 

G nah 
ad s 

W ta 

3 ti 

£ tla 


The Cherokee Syllabary 

308 Writing Systems 

latter is not to be taken as a phonemic representation, but rather 
as a possible writing system of an alphabetic type. Each symbol 
represents a vowel and a preceding consonant. There is nearly 
one-to-one correspondence between the pronunciation and the 
writing. Some departures, as provision of a symbol for ka beside 
a symbol for ga must be taken as evidence of influence from English 
where the voiced : voiceless contrast is phonemic, as it is not in 

21.12 True syllabic writing systems are not common. A few 
ancient, and mostly poorly understood, systems are thought to 
have been syllabic. The recently deciphered Minoan B was syl¬ 
labic. In recent times several other preliterate peoples have de¬ 
veloped syllabaries, in much the same way as did the Cherokee, 
getting the idea of writing by symbols representing sounds from 
some written language associated with Western civilization. In a 
few cases, notably Cree (Canada) and Miao (southwestern China), 
missionaries have designed syllabaries as a means of reducing 
languages to writing. None of these, however, have been of more 
than local importance, and many have been abandoned in favor of 
alphabetic writing systems. 

The one exception to this generalization is the Japanese syl¬ 
labary. Since Japanese is a large and important language spoken 
by a highly literate people, its writing system is of major impor¬ 
tance. The development of the syllabary antedates, and is inde¬ 
pendent of, contacts with Western culture. It is derived ultimately 
from the adaptation of the Chinese characters to the Japanese 
language. It is, therefore, necessary to return to the consideration 
of the former and examine the way in which they can be adapted 
for use with a language other than that with which they originated. 

21.13 China is today, and has long been, a country in which a 
number of languages (by local convention usually called "dialects,” 
though they differ as widely as the languages of Europe) are 
spoken. Many of them do, however, have certain basic similarities, 
w’hich are produced in part by common origin and in part by com¬ 
mon civilization. The result is that the morpheme inventories can 
be more or less matched up. On this basis, with exceptions of minor 
significance, any morpheme in any Chinese language can be written 
by using a character which a neighboring language would use for a 
morpheme of the same meaning. Thus the grapheme >». is used in 

Writing Systems 309 

each language for some specific morpheme having the meaning 
‘man.’ The morphemic reference will of course be pronounced quite 
differently in some of the languages: in Peking Mandarin [/ran], 
in Canton C\japl in Hakka [-pin], in Suchow Q/nen], in Fuchow 
[Miaq], in Amoy [v/laq], and in T'ang Min [\/t$in]. The char¬ 
acter k. ‘fire’ has reference in each of the same languages pro¬ 
nounced [/xwo /fo Mo Mmu -hui /he /ho]. This produces the 
peculiar result that documents written in different language areas 
can be read in any part of China, though the spoken languages are 
completely unintelligible to persons from another area who are not 

Much the same thing is true in Europe of those graphemes which 
are morphemic in reference. Thus 2 + 8 = 5 may be read in 
English two plus three is five, in German zwei und drei ist funf, in 
Italian due e tre fanno cinque, and in each other language in 
accordance with the vocabulary and grammar of that language. 

21.14 Such adaptation does impose some special requirements 
on the writing system. It is not possible to match up the morpheme 
stock of all the various Chinese languages completely. Each of the 
languages other than Mandarin will require some changes. The 
grapheme is needed in Cantonese to write the meaning ‘have 
not,’ but is not used in Mandarin. Some characters are used differ¬ 
ently in different languages. For example Mandarin writes 4* & 
to represent its two-syllable morpheme for ‘banana,’ while Can¬ 
tonese, having a one-syllable morpheme for the same meaning 
writes only There are also differences of order and of style, so 
that the written language is not entirely uniform over all of China. 

21.15 With the spread of Chinese learning, this same method 
of writing has been carried outside of China into areas in which 
the language structures are quite different. In every case the same 
common factor of Chinese cultural influences has operated to 
reduce certain of the difficulties. Annamese, Korean, and Japanese 
have all been written in the same way, by using Chinese characters 
for those morphemes which have translation equivalence to the 
original reference of the character, or for Chinese loan words. 
The difficulties of adaptation are, of course, much greater when 
the grammatical structure of the language is radically different 
from that of Mandarin Chinese. 

In particular, Japanese is a highly inflected language with 

310 Writing Systems 

numerous and often complex affixes. By contrast, most Chinese 
morphemes are roots, though a few affixes do occur. This means 
that no Chinese equivalents can be found for many commonly 
occurring morphemes. Various makeshifts were tried before a satis¬ 
factory method was elaborated. There are, in general, two alterna¬ 
tives: Additional morphemic signs could have been invented to 
represent the affixes as such, in much the same way as new char¬ 
acters have been developed to represent morphemes needed in 
colloquial Chinese languages. Or, signs of phonemic reference could 
have been added to the system. The latter course was taken, and 
a syllabary was developed out of patterns found within the Chinese 

The Japanese syllabary has developed in two forms, kalakana 
and hiragana. These are practically merely alternative written 
shapes of the same system. The basic graphemes are given in the 
following table. There are several additional features of the 
system that need not be described here. 

7 *> 









ii a' 




fit ke 

■a ; 







jr z 





ft te 

l- k 



ji. Iz 




/ o> 




















7 * 


D 9 









Katakana and Hiragana 

21.16 The Japanese syllabary was developed primarily as an 
adjunct to the Chinese characters, and is generally so used today. 

Writing Systems 311 

The roots are typically written in Chinese characters (known in 
Japanese as kanji), and the affixes in hiragana or katakana. Fre¬ 
quently the pronunciation of the kanji is indicated by writing 
small hirasjana symbols beside it. Alternatively, Japanese can be 
written wholly in one of the syllabic scripts. The difficulty with 
this is that the written style has numerous homophones, so that 
ambiguity may result. But if a colloquial style is adhered to, no 
such trouble should arise. That is to say, hiragana or katakana is 
reasonably adequate to represent spoken Japanese, but not satis¬ 
factory for the highly specialized literary written language. 

21.17 Alphabetic writing systems are those in which the graph¬ 
emes typically have reference to single phonemes. We must say 
"typically” because greater or lesser departures are almost uni¬ 
versal. Ideally, an alphabetic system should have a one-to-one 
correspondence between phonemes and graphemes. That is, each 
grapheme would represent one phoneme, and each phoneme would 
be represented by one grapheme. This condition is approximated 
(in some cases extremely closely) in the phonemic transcriptions 
of linguists, but in practical alphabets only to a limited extent. 
There are almost always other non-linguistic factors which must 
be taken into account when a new alphabetic writing system is to- 
be designed. Older systems may not have been phonemic to begin 
with, but commonly become less so as linguistic change operates 
on the spoken language. The one-to-one relationship is chiefly 
useful as a point of departure in discussing the fit of writing systems 
to spoken languages. 

The discussion in the remainder of this chapter will concern 
itself with some of the problems of fit of alphabetic systems to their 
spoken languages. Much of what is said applies in much the same 
way to other types of writing systems. 

21.18 English has been shown to have a phonemic system in 
which a number of sub-systems can be distinguished. Every other 
language probably has some similar phonemic sub-systems, though 
we cannot expect that the divisions will parallel those of English 
in detail. Alphabetic writing systems apparently are always re¬ 
stricted to the representation of selected sub-systems of the 

For example, the apparent ancestor of most of the alphabetic 
writing is that developed for the Phoenician language. This con- 

312 Writing Systems 

sisted of 22 graphemes. At the time there were apparently 22 con¬ 
sonant phonemes constituting a relatively well marked sub-system 
within Phoenician phonology. Within the limitation that only this 
one sub-system was represented at all, the Phoenician writing sys¬ 
tem was as close to being phonemic as any writing system ever is. 
There was no notation whatever for vowels or any other sub-system 
of the phonology. 

The oldest form of the Greek alphabet gave essentially phonemic 
representation of the vowels and consonants. It did not, however, 
give any indication of the pitch system, which we know to have 
been phonologically important. A system of diacritics to represent 
the pitch system was developed much later, and is customarily 
included in the modern conventions for the writing of ancient 

21.19 The Phoenician alphabet has been denied the status of a 
true alphabet because of its failure to write vowels. By the same 
token, the classical form of written Greek would have to be denied 
alphabetic status because of failure to indicate pitch contrasts. 
Greek pitches and Phoenician vowels are alike in constituting 
separate phonologic sub-systems from those represented in the 
writing system. Even the modern conventions of spelling ancient 
Greek would fall short, since there is every reason to believe that 
the phonemic system included more items than have been recorded. 
In particular, vowel length is known to have been phonemic, 
though it is only partially recorded by the accident that the vowels 
/te /rj and /o/co occurred only long. 

No alphabetic writing system is known which represents every 
sub-system in the phonology of the spoken language on which it is 
based. Such a system would unambiguously represent anything 
which could be said understandably. As we shall see this is not a 
good statement of the function of a writing system, which is rather 
to record a written language. The latter is different in certain ways 
from a spoken language. Moreover, it is doubtful that an alphabet 
which did accurately record speech would be practical. However, 
since almost no experimental work has been done on the design 
of writing systems, one cannot make any such statement cate¬ 

21.20 The problem of alphabet design, as far as our present 
empirical notions take us, would seem to include determining what 

Writing Systems 31 3 

is the phonemic system of the language, what are the sub-systems 
into which it is organized, and which of these are advantageously 
represented in the orthography. This latter question is most fre¬ 
quently raised in connection with pitch systems of so-called “tone 
languages” (in which the pitch phonemes are components of the 
commonest type of morpheme). In these, pitches usually constitute 
a sub-system, but one far less distantly removed from the con¬ 
sonants and vowels than is the case in a language like English. In 
some such languages, orthographies both with and without mark¬ 
ings for pitch have been tried. Results, as far as they are available, 
are contradictory. In some languages, pitch marking seems to con¬ 
tribute greatly to the usefulness of the orthography. In others, it 
seems to have little practical value, and native users tend to omit 
the pitch marks in writing, apparently largely disregarding them 
in reading. We can only guess at the controlling factors. 

The question is seldom raised as to whether other parts of the 
phonology need to be written. This is undoubtedly due to the 
dominance of the traditional pattern of European orthographies, 
in which consonants and vowels are written, with the addition of 
partial indication of stress in some cases. Were more native 
speakers of Semitic languages engaged in setting up orthographies, 
the question of writing vowels might be raised more frequently 
than it is, with results that no one can predict. Many languages 
have been written in orthographies patterned on Arabic, and gen¬ 
erally they have had rather loose indication of vowels. It is hard 
to say what part of the reported inadequacies of these writing 
systems is due to prejudices of European observers, to other short¬ 
comings of the Arabic script, or to the lack of thorough indication 
of vowels. Certainly, the question would be worth raising as to 
whether indications of consonants and vowels; or consonants, 
vowels, and pitches; or consonants, vowels, and stresses are the 
only types of alphabetic writing systems that are useful. 

21.21 A second type of departure from one-to-one relationship 
between the phonology and the writing system is due to conformity 
to other writing systems. This arises in several ways. Most com¬ 
monly it is the result of borrowing not merely a set of symbols, but 
also more or less of the rules of fit; and also it results from a con¬ 
scious or unconscious disinclination to take sufficient liberties with 
the system. A very early example is the adaptation of the Phoeni- 

314 Writing Systems 

cian writing system to the kindred dialect of Hebrew. In spite of 
close relationship, Hebrew had more consonant phonemes than 
Phoenician. We do not know exactly how many, since certain of 
them were disappearing at about this time. One that is thoroughly 
attested can be written as /*§/, though we do not know its exact 
phonetic value. This was written with the same character as that 
used for Phoenician and Hebrew /*§/• Presumably, this was the 
“nearest equivalent” in some rough sort of way. In any case the 
Hebrew writing system lost the one-to-one character which 
the Phoenician prototype had, with consequences that will be 
discussed in 21.25. 

21.22 English orthography fails in many respects. These go 
back to many origins, but among them is a too slavish following 
of the Latin writing system. The Latin alphabet, rather close to a 
one-to-one representation of the consonant and vowel phonemes, 
consisted at one time of 21 letters. Within the period when Latin 
was still a living language, two more, y and z, were added to pro¬ 
vide for new phonemes introduced into the language in Greek loan 
words. Three others, j, v, and w are medieval or modern additions. 
J and v have been read back into the classical texts, whence we 
read Vent, vidi, vici for an earlier uenj uidi uici. (The contrast of 
capitals : lower case is modem.) English-reading people have stub¬ 
bornly resisted any further additions. This is in spite of the fact 
that an older writing system for English (written Old English or 
“Anglo-Saxon”) used a few other letters, including S and £ (6). 
The retention of these and the addition of several others would 
have provided a much better base for a practical English orthog¬ 
raphy. Some other European nations have been a bit less con¬ 
servative than English. Norwegian has added ze p d; Spanish added 

and considers U and rr as single alphabetic units; and similar 
small departures have been made in other languages. But most 
European writing systems include all 26 of the “Latin Alphabet,” 
whether they are all used or not. In general, the alphabet has been 
considered something not easily tampered with. This attitude has 
been carried out of Europe in recent years by some missionaries. 
(The vast majority of new writing systems have been and are being 
designed by missionaries.) One even claimed as the chief virtue of 
the orthography he had created, that it employed all twenty-six of 
the letters and no others! 

Writing Systems 315 

21.23 In recent years, and as one of the important products of 
newer understanding of linguistics, there has been a tendency 
toward freer use of the Latin Alphabet. This includes a greater 
willingness both to discard unneeded symbols and to add others 
as required. The conspicuous example is the development of the 
“Africa Orthography,” a series of flexible recommendations for 
the design of new writing systems for African languages. The result, 
when well applied, has been increased simplicity, readability, and, 
frequently, improvement in appearance. 

21.24 Conformity to some other writing system, even at the 
expense of a one-to-one relationship, may sometimes be desirable, 
especially in areas where bilingualism is common. Thus it has fre¬ 
quently been judged necessary or desirable to make orthographies 
conform to that of the dominant or governmental language in the 
area. This does not always sacrifice the one-to-one relationship, 
even though at first sight it would seem to do so. For example, in 
some Spanish areas a phoneme /k/ may have to be written c before 
aou and gu before i e, following a similar convention in Spanish. 
But since c and gu are in complementary distribution, they may be 
considered as allographs of one grapheme, and hence the one-to-one 
relationship is preserved at the grapheme level, though not at the 
allograph level. 

21.25 Purely orthographic troubles, as well as deeper problems 
to be mentioned in the next chapter, are sometimes the result of 
historical change in the phonology of the language. Consider again 
the Hebrew writing system mentioned in 21.19. At the time of the 
establishment of the writing system there were three phonemes 
/*s/, /*6/, and /*§/ with which our example is concerned. The first 
was written (s). The second and third were both written as (§). 
In the course of time the contrast between /*s/ and /*§/ was lost. 
There were now two phonemes /s/ and /§/. The difficulty is that 
now one phoneme /s/ is written two ways, (s) and (§), and one 
grapheme (&) has two phonemes as reference, /&/ and /§/• The 
situation was not satisfactory at the start of the development, but 
it was worse at the end. 

The results of change are not always so undesirable. There is 
some reason to believe that the grapheme ( < i) was used to repre¬ 
sent two phonemes, /*V and /*y/. But the historical develop¬ 
ment (probably completely independent of the writing system) 

316 Writing Systems 

was for these two to fall together into one phoneme /VThe result 
was that what had been a failure of the one-to-one relationship 
was remedied by the fortuitous course of phonologic change. 

Nor is it necessary that there be a confusion such as that between 
/*§/ and /*§/ to begin with. The oldest form of Hebrew apparently 
had a phoneme /*b/ which was written (b). At a somewhat later 
date this had developed two quite distinct allophones [b] and [v], 
but as they were in complementary distribution and hence still 
allophones of one phoneme, the relationship with (b) was still 
one-to-one. In modern Hebrew, however, they are no longer in 
complementary distribution. The old /*b/ has been replaced by 
two phonemes /b/ and /v/, but the writing system still contains 
(in one form) only one grapheme (b), and the relationship has 
become two-to-one. 

21.26 If a writing system does not represent the whole pho¬ 
nemic system, as we have seen none do, then it follows that written 
representation of spoken material may be less clear than the speech 
it records. This is of course true in English, as we pointed out in 
10.16. Written language usually compensates for this by various 
devices, most of which fall within the topic of the next chapter, but 
some are part of the orthography or writing system. The most 
familiar of these is word division, a sort of crude morphemic 
signal. Its effectiveness may be realized by trying to read material 
in what is sometimes called a “phonemic transcription,” but with¬ 
out indication of stress, pitch, transitions, or terminals. With word 
divisions given, such material can be read with no great difficulty; 
without word divisions it can be exceedingly obscure. 

Another device is arbitrary spelling distinctions. This device may 
separate words brought into homonymy by omission of part of 
the phonemic system from the base of the orthography. Latinxua, 
an alphabetic writing system for Chinese, does not indicate the 
pitches. It is believed by the designers of the system that only a 
few pairs of words can be confused because of this. These pairs 
have been distinguished by artificial, arbitrary spelling devices 
which are not used generally in the system. Thus liz ‘chestnut,’ 
mat ‘sell,’ yarn ‘courtyard,’ and Shansi ‘Shansi’ arc given the 
spellings which the rules would suggest, whereas liiz ‘plum,’ maai 
‘buy,’ yaanz ‘garden,’ and Shaansi ‘Shensi’ have spellings ar¬ 
bitrarily set up to be distinctive. Incidentally, it is to be noted 

Writing Systems 317 

that the last two, the provinces Shansi and Shensi, are accidentally 
distinguished by a similar device in English spelling, the usual 
Western pronunciation being derived from the spelling. 

Though not so purposefully planned, numerous pairs of words 
in English are distinguished by arbitrary spelling differences: 
hole : whole, lead /16d/: led, boy: buoy, etc. The virtues of such 
distinction are sometimes overlooked by advocates of spelling re¬ 
form. It may well be that, short of a sufficiently drastic change, 
“correction” of a few of the most glaring cases on a haphazard 
basis may actually make for greater difficulties in the use of the 
English writing system. The design of an orthography (spelling 
reform is merely a small-scale attempt) is a difficult and intricate 
matter about which we as yet know all too little. 




Written Languages 

22.1 A written language is basically a representation of a spoken 
language. It is, however, very seldom an exact reflection. Mention 
was made in the last chapter of some of the ways that a writing 
system might fail to represent the phonology fully and accurately. 
These are, however, only special instances out of many other 
differences — often much deeper differences — between a written 
and a spoken language. These differences are found at all levels 
of structure — phonology, morphology, syntax, vocabulary, and 

22.2 Dialectal variation in spoken languages is a very familiar 
phenomenon. Dialectal variations are also found in written lan¬ 
guages, but they are generally minor and not at all obvious. In 
English a few are well known: colour : color, gaol: jail, com : grain, 
the government are . . . : the government is . . ., etc. Differences be¬ 
tween the spoken English of Britain and the United States are 
considerably more numerous and often much greater. 

This situation is very nearly universal. Dialectal variation in a 
written language is almost always much slighter than in the asso¬ 
ciated spoken language. Sometimes speech differences may be so 
extreme that there is no mutual intelligibility, whereas the written 
language in the two areas is identical. 

22.3 An illustrative case is that of German. Over most of 
Germany and Austria, in a large part of Switzerland, and in small 


Written Languages 319 

portions of other European countries, the people consider their 
language to be German. They are, of course, aware that not all 
spoken German is alike, and they may be conscious of considerable 
differences between their own and other speech also called German. 
But even when these differences are extreme, they consider the 
difference as merely dialectal. Many of these “dialects” are com¬ 
pletely unintelligible to speakers of other dialects of German. The 
differences are far greater than can be found in the United States 
or among the more familiar dialects of Britain. Some are more dis¬ 
tinct than the different Scandinavian languages, Swedish, Danish, 
and Norwegian. 

Yet over all this region with its tremendous speech diversity, 
there is one universal written language with only very minor 
dialectal variations. As a result, any literate person can com¬ 
municate with any other in writing, or by a spoken rendition of 
written German, which is a very important special dialect of quite 
different status from the local colloquials. It is often known as 
“Schriftdeutsch.” Every school child is taught to read, write, and 
speak this common literary language. For many of them the task 
is little different from that of learning a second language. If 
American children were taught to write and speak Schriftdeutsch 
and to consider their spoken English as merely a German dialect, 
their situation would be only a little more extreme than that of 
some German and Swiss children. 

Obviously for many people in the German area there may be 
very great differences between their speech and the written lan¬ 
guage they most commonly use. In some parts of Switzerland, for 
example, the colloquial uses /ksiy/ ‘been.’ In a more formal situa¬ 
tion, as in school or church, Schriftdeutsch would be spoken, and 
/govenzan/ would be used for ‘been.’ In writing, gewesen would 
be used, not only in all formal situations, but in most informal as 
well. That is, gewesen would commonly be written where one 
might say /ksiy/. 

In English we have something of the same thing, but it is very 
much less extreme. For example, I commonly use /ytiwonl/ as a 
plural pronoun contrasting with the singular /yuw/. This is re¬ 
stricted to colloquial situations. When speaking formally I use only 
/yuw/ for either singular or plural. In writing I use standard Eng¬ 
lish you almost exclusively, just as a Swiss would use standard 

320 Written Languages 

German. Not infrequently when reading aloud I say /ydwoHl/ 
when the context demands it, though you is written. The difference 
is that such cases are comparatively rare in English, but very 
common in some dialects of German. 

22.4 Written German had a long and gradual development. It 
was originally based on a Middle German dialect. By historical 
accident its use spread over most of the Middle and High German 
dialect areas and over a large part of the Low German dialects. It 
successfully displaced most of the other written languages which 
had started developing in the area. Thus there came about a 
remarkable degree of uniformity in written language over a large 

Two other written languages were, however, successful in re¬ 
sisting this spread. This was largely the result of differences in their 
political situation. Each spread throughout the Germanic area of 
the political unit in which it arose. Thus we have today Dutch in 
the Netherlands and Flemish in Belgium. It is not surprising that 
this should have happened, since common written language is an 
important factor in integrating a political nation, and since the 
unity within encourages the spread of any culture pattern through¬ 
out a major political unit. The unique element in the German 
situation is that the common written language could have spread 
beyond one national state, and that it should have produced as 
little irredentism as it has. 

The spoken bases of Dutch and Flemish were very similar. The 
two written languages were accordingly very similar, and the trend 
in each has been toward conformity with the other. As a result 
these two have become almost identical, differing little more than 
British and American written English. 

The spoken dialects of the Netherlands and adjacent parts of 
Germany were practically identical. The written languages used in 
the two areas are, however, quite different. The geographic and 
social limits of written languages are not necessarily correlated 
with those of spoken languages. This is only one case of many of 
similar discrepancies which can be cited both in Europe and else¬ 

22.5 In some instances the origin of the written language is 
quite clear. In many others it is very obscure or involved. Let us, 
however, assume the simplest case: A new written language is 

Written Languages 321 

developed based on one single dialect, A. After it has been reason¬ 
ably well established its use spreads to adjoining areas using related 
spoken dialects. As its area of use widens it comes in contact with 
ever more divergent dialects, and ultimately with dialect B, which 
differs markedly from dialect A. Speakers of B will learn to write 
this written language, and so to use some of the vocabulary and 
grammatical patterns of dialect A, many of which are unnatural to 
them. It is almost inevitable that they will introduce into their 
writing some vocabulary and perhaps some grammatical usages of 
their own speech. What speakers of B write is not, therefore, 
identical with the original written language. Instead it is modified 
somewhat in the direction of B. Dialect differences, perhaps only 
very minor, have appeared in the written language. 

Some of these new usages will be taken up by writers from other 
dialect areas, thereby reducing the distinctiveness of the written 
dialect in the area of B. Items originating in dialect B may, thus, 
spread throughout the written language. Often they will become 
alternatives existing alongside usages originating in dialect A or 
similarly introduced from other dialects. One effect is to enrich the 
vocabulary and grammatical repertoire of the written language. 
This is a common source of the synonyms which are so abundant 
in some written languages. 

A second and ultimately more important effect is to produce a 
new literary language which is not merely a reflection of any single 
dialect, but a composite of many. If this process is continued, 
the point may be reached where there are appreciable differences 
between written and spoken language for all dialects. 

22.6 Modern literary Italian is an excellent example of this 
process. This can be traced rather definitely to an origin based on 
the speech of Tuscany. Dante’s Divina Commedia was the first 
significant publication in Italian. It had an immediate and deep 
influence on the development of the written language, and that 
influence has continued. Dante based his work on the speech of his 
native Tuscany. A large proportion of the other early writers of 
importance were also Tuscans, and the region was for some time 
the cultural center of the Peninsula. 

This written language was gradually adopted throughout the 
area which later became Italy. It came to be associated with 
dialects which were markedly different from that on which it had 

322 Written Languages 

been baaed. Some authors from remote areas succeeded in writing 
in very accurate imitation of their Tuscan models; others intro¬ 
duced numerous elements from other Italian dialects. Some of these 
were in turn imitated in Tuscany and elsewhere. With the rising 
importance of Rome as an Italian cultural center, it was inevitable 
that many usages from that area should be introduced and spread. 
But no dialect area failed to make its contribution. Modern written 
Italian is not provincial, but the common language of the whole 
country. In many respects it is a sort of average of all the local 

22.7 A common written language such as German or Italian 
is a powerful unifying force in a national or cultural area. It is also 
a necessity if a flourishing literature is to be developed, since it is 
not economically feasible to develop a separate written literature 
for every minor local dialect. In many parts of the world todaj', 
adequate written languages are badly needed in order that the 
peoples concerned may take their rightful places in the modern 
world. No nation can afford the long, slow process of development 
by which the written languages of Europe mostly came into their 
present situations. It is desirable to short-cut this slow growth by 
some quicker way to an adequate written language usable by a 
sufficiently large population to support a vigorous literature. 
One device has been the deliberate design of so-called “union” 

The Shona group of dialects occupies a large part of Rhodesia 
and adjacent Portuguese East Africa. There are six reasonably 
clearly marked groups of dialects, each showing appreciable local 
variation. Five of these were reduced to writing by missionaries, 
and in four of them the whole New Testament was translated. An 
appreciable amount of publication was done in each. However, all 
the Shona dialects together are spoken by only a little over a mil¬ 
lion people. This is hardly enough to support an adequate output of 
printed materials in five different written languages. Accordingly, in 
1929 a committee began work on a survey of the language problems 
in the area. As a result of their work, a new written language known 
as Union Shona was designed. It has since largely replaced the 
older written forms and is now quite generally used throughout 
five of the six dialect groups. The committee decided that it was 
impractical, for reasons as much social and geographical as lin- 

Written Languages 323 

guistic, to bring the Kalanga dialects into the scheme. A unified 
grammar was set up, based largely on the Karanga and Zezuru 
dialects. The vocabulary was drawn largely from four of the 
dialects, and it was agreed to discourage the introduction of new 
words from others. A new and improved spelling system was intro¬ 
duced. For example, c is now written instead of the older ch, and 
five new letters were proposed to write phonemic distinctions 
previously overlooked. 

Since that time Union Shona has become well established. There 
has, of course, been opposition, both in general and to specific 
features, but it has not been serious. Numerous publications have 
appeared, including the entire Bible. The scheme is proving its 
value in many ways. Not only has the course of development been 
greatly speeded, but a natural group of dialects has been selected 
for inclusion. In time the artificiality of the design will become less 
obtrusive and a highly satisfactory written language may be ex¬ 
pected to develop. 

22.8 Such union languages are not always equally successful. 
Union Ibo for a large and linguistically very diverse area in 
southern Nigeria had a very different outcome, and has been 
largely abandoned. The failure seemed to be due to poor selection 
of the features to be included, and to inadequate attention to the 
social and political differences in the area. That is to say, there was 
inadequate field investigation both in linguistics and anthropology 
on which to base the proposal. 

This and other experience has shown that the successful design 
of a written language requires thorough knowledge of the linguistic, 
social, political, and practical aspects of the problem. Hence it 
requires cooperative work by linguists, anthropologists, adminis¬ 
trators, educators, missionaries, and native leadership. It is one 
of the very practical problems to which descriptive linguists and 
dialect geographers can make important contributions. 

22.9 Written languages not only are influenced by the spoken 
languages with which they are used; they are also influenced by 
other written languages. This is most evident in their early de¬ 
velopment. Generally the first authors in any newly written lan¬ 
guage are bilingual. Often they are only secondarily speakers of the 
language being “reduced to writing.” In the worst case their 
command of the language may be poor. But even when the author 

324 Written Languages 

is a native speaker, his literary education has usually been wholly 
through the medium of another language. His understanding of the 
nature and possibilities of written language is based on the writing 
system and literary grammar employed elsewhere. It is almost 
inevitable that certain conventions of the pattern system will be 
carried over, irrespective of their suitability. 

Foreign influence does not cease after the literary conventions 
of the new language are firmly established. Authors are much more 
likely to be bilingual than are the general public. They sometimes 
make a conscious effort to imitate models from other literatures. 
Still more frequently they arc unconsciously influenced by literary 
patterns in other languages. Translations also play an important 
part in this process. It is rarely possible to translate without some 
conformity, often considerable and fundamental, to the language 
structure of the original. The patterns so introduced may be copied 
and in time be thoroughly assimilated into the written language. 
In many languages the Bible is among the first publications; this 
has been an important fact in bringing a certain measure of com¬ 
mon usages into many written languages. 

22.10 The English writing system includes a number of features 
other than the alphabet. One prominent element is the punctua¬ 
tion system. For example, consider the marks of sentence ends, 
(.) (?), etc. In the first place, a “sentence’ 1 is regarded as “ex¬ 
pressing a complete thought,” or something of the kind. The 
writing system requires that it must have a terminal mark and 
begin with a capital letter. In the second place, “sentences” are 
classified on the basis of whether they “state a fact,” “ask a 
question,” etc. These differences determine which of the several 
terminal marks will be used. Somewhat comparable structures are 
marked in spoken English by the use of the intonation contours. 
The two systems are, however, quite independent. Sentences which 
would be written with (?) are clearly divided in speech into two 
groups with sharply contrasting intonations. One of these uses 
patterns which arc also commonly used with sentences which 
would be written with (.). (Cf. 4.18.) There is, therefore, only 
more or less incidental correlation between the punctuation marks 
and the different intonation contours. Indeed, intonation contours 
commonly end at places where punctuation marks are prohibited, 
as between certain subjects and the following predicates. 

Written Languages 325 

There is a fundamental difference in attitude toward the two 
parts of the writing system. The alphabet is assumed to represent 
sounds. Everyone expects, therefore, that there will be differences 
of spelling from language to language. The punctuation marks, 
however, are not conceived of as representing features of speech. 
Instead, they mark logical units of connected writing like “sen¬ 
tences,” “statements,” “questions,” “dependent clauses,” and 
the like. Since these are phrased in terms of logic, they are gen¬ 
erally assumed to be universal. People do not expect to find differ¬ 
ences of punctuation from language to language. Indeed, they are 
often irritated by the very minor differences which they do find 
between English and, say, French or German. Because the punc¬ 
tuation system of all European languages has grown up through 
mutual interaction and on the basis of a common “logical” 
grammar, all are basically similar. Certainly, they are far more 
similar than the intonation systems in the corresponding spoken 

22.11 In recent years a very large number of languages of Asia, 
Africa, and America have been “reduced to writing,” usually by 
Europeans and Americans. Generally an attempt has been made 
to adapt the alphabet to the sounds of the language. But punctua¬ 
tion and capitalization have in most cases been carried over with 
little or no change. For example, proper names are usually cap¬ 
italized, whether this has any functional value in the language or 
not. First words of sentences are capitalized, though this is strictly 
redundant, even in English, since sentence ends are also marked. 
Questions are carefully distinguished by (?) though in many 
languages all questions are clearly marked by special words. 

The significant thing is not so much that these conventions are 
carried over, as that no questions are commonly raised about them. 
This is, of course, almost inevitable if they are conceived in terms 
of logical categories which are more or less universal. 

No one knows what would be the outcome if thoroughgoing 
study were applied to problems of this kind. There may well be 
far better methods of indicating syntactic relationships in written 
language than any which have been developed hitherto. No appre¬ 
ciable amount of research has ever been devoted to this question, 
though a much larger amount of attention has been given to the 
problem of selecting alphabets and spelling conventions for new 

326 Written Languages 

written languages. Indeed, there is available very little descriptive 
data on how the English, or any other, punctuation system is 
actually used. The large volume of published material which is 
available is predominantly normative and almost wholly based on 
'‘logical” categories. 

22.12 We must not consider the question of punctuation for a 
new written language as a trivial problem. Punctuation marks 
syntax. It is predicated on a certain type of syntactic structure. 
The patterns used in the spoken language may be very different. 
If European punctuation is introduced, there must be pressure in 
the direction of European syntax. At the minimum this may do 
nothing more serious than to favor certain perfectly natural con¬ 
structions at the expense of others. In some cases it has, however, 
introduced previously unknown patterns. Of course, the question 
is much more extensive than we have framed it. Punctuation, 
capitalization, and word division are only the most obvious 
features of a much larger set of patterns. Various elements of 
syntax are carried over from one language to the next, often with 
the result that the written language departs in significant respects 
from the spoken language on which it was based. 

22.13 A third factor influencing the fit of written and spoken 
languages is linguistic change. The picture presented up to this 
point has been largely static. Actually, the spoken dialects are 
relatively rapidly moving entities. Change is taking place con¬ 
tinually in every aspect of every dialect, though, of course, not all 
of these changes are independent of those in other dialects. More¬ 
over, written languages are also subject to change beyond the 
broadening of the base which has just been described. The prob¬ 
lem then can be stated in terms of the comparative rates and 
directions of change in the written language and the spoken 

The most obvious instance is the changes in phonology and the 
relatively much slower change in orthography. Almost inevitably, 
the fit of orthography to pronunciation deteriorates. This is, of 
course, a major source of spelling difficulties in English. But though 
orthography is notoriously conservative, it must not be assumed 
that it is immutable. A letter-by-letter transcript of the first edi¬ 
tion of the King James Version (1611) will show the actual extent 
of change. 

As printed in 1611 : 

And hee said, A certaine man had 
two sonnes: And the yonger of them 
said to his father, Father giue me 
the portion of goods that falleth to 
me. And he diuideth vnto them his 

Written Languages 327 

As in modem printings: 

And he said, a certain man had 
two sons: and the younger of them 
said to his father, Father, give me 
the portion of thy substance that 
falleth to me. And he divided unto 
them his living. 

22.14 Spelling reform in many languages has been, as in Eng¬ 
lish, largely a private matter. Certain individuals have flouted 
public opinion and made changes in their own spellings. Whether 
such innovations prevail depends on public reaction, and this is 
a wholly unpredictable matter. Some have been accepted by in¬ 
creasing numbers of people until the new spelling has come to 
dominate. Thus, in America jail has replaced gaol, though the 
latter remains in Britain. Many others have failed of general 
acceptance. A number of proposals are currently competing with 
older spellings, often generating appreciable controversy, as in the 
case of nite : night. With such a haphazard process it may be ex¬ 
pected that there will be no consistency in the directions of change. 
Most of the changes are probably to be considered as in the direc¬ 
tion of simplicity and better fit, but this is not always the case, 
as when rhyme supplanted an older rime. 

In some other areas, notably Scandinavia, the initiative is 
largely official. For example, the Ministry of Education in Den¬ 
mark recently decreed that all old spellings in aa would be replaced 
with d. No Dane is compelled to make the change, but the schools 
will now teach the new form. Swedish orthography has had a 
reasonably thorough revision about once each generation. This 
has not, however, insured that spelling will accord with pronun¬ 
ciation, since a number of phonemic distinctions are still unre¬ 
corded, and unpronounced letters, though diminishing in number, 

22.16 The effects of phonologic change on the fit of a writing 
system are various. Of course, a phonetic change without accom¬ 
panying phonemic change has no effect. Phonemic changes may 
merely produce an alternative but unambiguous orthographic 
device. For example, light was at one time pronounced /lixt/; gh 
was the usual spelling for /x/. Through a series of changes /lixt/ 
became /l&yt/. But since exactly parallel changes affected a large 

328 Written Languages 

number of other words, the result was primarily to add igh to the 
list of possible spellings for /&y/. Thus there can be little confusion 
in reading; light can only spell /I&yt/. The converse is not true; 
one must learn to spell light and not Hite. The patterns of phonemic 
change in English since about 1400 have been so complex and have 
so often been further influenced by analogic changes that the 
consequences are not always as simple as this. 

Some aberrant spellings have another function within the struc¬ 
ture of written English. Consider the pair sight and site. It is quite 
clear and correct to write The sight is pleasing, or The site is 
pleasing. That these two words are homophonous docs not affect 
their usability within the written language. But neither of these 
sentences is used in normal speech. Instead, something like It's a 
nice view, or I like the location, are much more likely. The ultimate 
effect of these spellings is to provide a semi-morphemic writing 
system for the language. Within the limits of the alternative 
spellings possible, one is selected to act as a marker of the specific 
morpheme meant. The difference in spelling between sight and 
light has a phonemic reference; that between sight, site, and cite has 
a morphemic reference. The English writing system is, therefore, 
only partially phonologic in its basis. 

22.16 Written and spoken languages differ in their vocabu¬ 
laries. The first impression is always that the long-established 
written language has a much larger vocabulary than any spoken 
dialect. In part, this is illusory. The general public, and even pro¬ 
fessional linguists, have notoriously underestimated the vocabu¬ 
laries of spoken languages. Colloquial words can easily be missed, 
as some arc under verbal taboos which prevent their use in writing 
entirely and in public speech partially. They may, nevertheless, 
be widely current in the community. Others apply only in situa¬ 
tions where written communication is unlikely, as, for example, 
technical terms in daily activities. (Few languages other than 
English have the vast literature on cooking which pours out each 
month from American publishing houses.) Others are unrecorded 
because more literary forms are preferred in speaking in the pres¬ 
ence of strangers. 

There are, however, two characteristics of literary vocabularies 
which seem to be sufficiently general to require comment. These 
are the relative abundance of homonyms and of synonyms. 

Written Languages 329 

22.17 Many written languages have large numbers of homo¬ 
nyms. This is not strictly an accurate way of stating the matter, 
since many of them are amply distinct in spelling, though they 
would be pronounced alike. Some of them owe the difference in 
spelling to the conservatism of orthography which we have just 
mentioned. English is only modestly equipped with such pairs. 
Literary Chinese, however, has very many more. It is quite readily 
possible to construct long passages which are quite intelligible to 
the eye, but utterly incomprehensible to the ear. 

Occasionally homonyms are artificially distinguished by some 
orthographic device. A somewhat unusual case in point is the 
English word calorie. Several definitions have been proposed, two 
of the common ones giving values differing by a factor of 1,000. An 
arbitrary convention was established of writing one Calorie and 
the other calorie. This is now little used, but does persist in the 
abbreviations Cal. and cal. A somewhat similar instance is the 
word species /spiysiys/, of which the singular and plural are alike 
both in spelling and pronunciation. Abbreviated, however, there 
is a distinction, sp.: spp. (The doubling of the final letter is a 
common written allomorph of {-Zj}.) Occasionally these abbrevia¬ 
tions are used in contexts where no other words would be abbre¬ 
viated, simply because they are less ambiguous than full forms. 
Moreover, they are occasionally used as glosses in written English 
species (sp.) : species (spp.). 

22.18 The second peculiarity of written vocabularies is the 
abundance of synonyms or near synonyms. Often these are words 
originating in different dialects. Or one or more may be loan words 
from some other language. In any case, once a word enters a 
literary language it tends to persist, since there is a certain per¬ 
manence to written records lacking in any spoken utterance. 
Moreover, in some literary traditions the repetition of a word is 
considered poor style except when a special effect is desired. This 
puts a special value upon synonyms. However, this feature of style 
is not universal. In some languages exactly the opposite effect is 
sought; repetition, especially following certain established pat¬ 
terns, is considered desirable. 

A third force favoring extensive groups of synonyms in some 
languages is the lack of indication of intonation and other qualities 
of voice associated with emotion. As a substitute, words of similar 

330 Written Languages 

denotation but diverse connotation can be employed. For example, 
a purely colloquial reporting of an extended dialogue will take 
the the form: “. . . So he said ... So he said . . .” The intervening 
utterances will carry intonations approximating those in the 
original, and so conveying the emotional tone. In a literary report 
said will be replaced throughout by verbs selected from a list in¬ 
cluding aslxd, exclaimed, shouied, snapped, growled, etc. Each of 
these, in effect, can be analyzed as carrying two functions: One 
(the denotation) is to replace said, the other (the connotation) is to 
portray the emotional tone of the utterance reported. 

22.19 All these forces may interact to produce very complex- 
relationships between speech and writing — how complex may 
best be seen in a situation where there are two competing written 
languages associated with the same spoken dialects. One such case 
is in northern India. One written language, known as Urdu, was 
developed in a Muslim cultural environment and uses the Persian 
form of the Arabic alphabet. The other, Hindi, was developed in a 
non-Muslim environment and uses the Nagari script which was 
borrowed from Sanskrit. These are more than merely two forms of 
writing for the same language. If Urdu is transcribed into Nagari, 
it is still recognizably Urdu. There are many other differences, 
and the two must be considered as more or less independent 
written languages, each with its own characteristic structure and 
vocabulary. The two differ in part because they were originally 
based on somewhat different spoken dialects. Each has spread over 
a very large and linguistically diverse territory. There are, how¬ 
ever, areas in which Urdu is rarely used, even by Muslims, but 
where Hindi is the prevailing written language. There are also 
areas where Urdu is used but not Hindi. Since they have not been 
used in identical territories, they have been subject to different 
influences from spoken dialects. The external influences have also 
been different. Urdu has been subject to influence from Persian, 
and this has affected every level of structure, not only vocabulary. 
In Hindi the Persianizing forces have been much weaker, but there 
has been a strong pressure for conformity to Sanskrit patterns. 
Probably most important of all, however, is the fact that each has 
developed more or less independently of the other. The historical 
changes which are inevitable in any language, spoken or written, 
have been different. As a result, Hindi and Urdu show important 

Written Languages 331 

and quite evident differences, and both are quite different from 
the spoken dialects of the area. 

22.20 Thus far, the discussion has centered around the struc¬ 
ture of written languages and their dependence upon spoken 
languages. This is only part of the picture. The interaction is in 
many respects mutual. As dialects contribute to a standard 
written language, the latter exerts forces on the dialect, producing 
convergence from that side too. The relationship is rather more 
complex than in the opposite direction for two reasons. The first 
is that usually only a small number of the speakers of the dialect 
are literate, and hence subject to direct influence, whereas almost 
all of the writers also speak some dialect. Moreover, the spoken 
language is likely to have a complex structure of levels of speech. 
The written language, will, of course, also have a set of literary 
levels, but these are probably less significant. 

One effect of a well established standard written language is 
the creation of a speech form which approximates an oral rendition 
of the literary language. This is familiar enough in English, where 
a certain type of oratory, now less popular than formerly, was 
scarcely anything else than a spoken written English, with a special 
kind of intonation substituted for the punctuation. From that, 
all intergrades can be heard with decreasing use of literary vo¬ 
cabulary and sentence structure, until a purely colloquial speech 
is reached. Actually, few educated Americans are able to speak 
without some evident traces of the literary language. 

22.21 The general influences of literary language on speech 
are probably much more significant than any of the more obvious 
details. But the latter are easier to point out, and collectively 
indicate a large group of changes. Among the more obvious are 
spelling pronunciations. For example, I usually read awry as 
/6Hriy/, though I know, when I reflect, that /oray/ is etymologically 
better. Solder is heard increasingly frequently with an /l/. Not all 
spelling pronunciations are substandard; some are very widely 
accepted. For example, use of /l/ in soldier is a recent development 
of this kind, as the word was earlier /sbwjor/. A very large per¬ 
centage of the rarer literary words, when pronounced at all, are 
pronounced as the spelling would indicate, irrespective of what 
oral phonologic tradition might have indicated. The same is true 
of more common colloquial words than is commonly realized. 

332 Written Languages 

Then there are such evident intrusions of written forms into 
speech as /ydwn&skow/ for UNESCO and /6nr&H/ for UNRRA. 
Most of our plural formations of Latin and Greek origin (see 
8.10.6) are preserved in speech, or enter speech in the first place, 
through the influence of written forms. The occasional American 
who consistently says /ky* §ffil . . ./, and even those who only occa¬ 
sionally use such forms, reveal the influence of writing, since this 
usage originated in written English. 




Language Classification 

23.1 Descriptive linguistics is concerned with two very different 
but intimately related tasks. The first is to describe individual 
languages or dialects in terms of their own characteristic structure. 
For each of the numerous speech forms this is a separate task; the 
structure of no other language is directly relevant. Linguists and 
all who make use of their results are naturally interested in know¬ 
ing how this work is progressing and how much remains to be done. 
This raises the question as to how many languages there are. 

The second task of descriptive linguistics is to develop a general 
theory of language structure — that is, to set up a conceptual 
framework within which an investigator can work as he seeks to 
understand a specific language. This theory must be sufficiently 
general and flexible to provide for any type of language structure 
that may be encountered, but also sufficiently precise and sys¬ 
tematic to give real help. This second task can only be accomplished 
by comparing numerous languages and language descriptions, and 
abstracting from them those patterns which seem of interest and 

Since languages are so numerous, no one can control this large 
mass of material without some sort of classification. The classifica¬ 
tion of languages is not within the province of descriptive lin¬ 
guistics, though many descriptive linguists also work on this 
problem. For descriptive linguistics a language classification is 


334 Language Classification 

simply a tool provided by another discipline. But it is a tool which 
cannot well be dispensed with. 

This chapter will, accordingly, consider some of the problems 
involved in the enumeration and classification of languages. The 
last chapter will give a broad outline of a language classification. 

23.2 One basic difficulty is that of defining a language. The 
word has been so variously used both by laymen and linguists that 
there is no agreed meaning, except for a genera! feeling that 
languages are somehow more distinct than dialects: languages are 
different kinds of speech; dialects are merely varieties of languages. 
Nothing as vague as this can be useful as the basis of a scientific 
enumeration of languages. 

It would be quite defensible if linguists would take it upon them¬ 
selves to redefine the term with sufficient precision that later 
linguists would be able to understand with some exactness just 
what is meant. Unfortunately, the problem involves a great deal 
more than definition of terms. The very nature of language is such 
that the problem of classification into such categories as language 
and dialect is intrinsically difficult or impossible. Several criteria 
can be proposed, no one of which is satisfactory. 

23.3 Perhaps the most obvious criterion is that of mutual 
intelligibility. We expect speakers of any given language to be able 
to understand each other. Conversely, speakers of different lan¬ 
guages cannot ordinarily understand each other. We might apply 
this as a criterion by inverting it: If two people can understand 
each other, then they speak the same language; if not, then they 
speak different languages. 

Unfortunately, it is not as simple as that. In the first place, 
intelligibility is a relative matter. Anything from essentially one 
hundred percent to zero may be found if such a test is applied. 
How much do two people have to understand of each other's speech 
to indicate that they speak the same language? Moreover, the 
matter is much affected by various complicating factors. For 
example, intelligibility depends on the subject matter. Try the 
following bit of English: “Stamens dimorphic; anthers oblong to 
subulate, truncate to attenuate or rostrate at the summit; con¬ 
nective of the larger anthers greatly prolonged and bearing two long 
basal anterior appendages, that of the smaller anthers much 
shorter, simple or merely bituberculate.” Intelligibility also de- 

Language Classification 335 

pends on the intelligence and background of the informant. The 
person with broader contacts, even if restricted to his own group, 
may have an appreciably higher comprehension of texts in a 
related speech form. 

23.4 Another criterion which is sometimes a good bit easier to 
apply is that of enumeration of common elements. For example, 
we may take a list of basic vocabulary and compare two speech 
forms. Such a list would include the ordinary everyday words 
for such meanings as: ‘father,’ ‘hand,’ ‘food,’ ‘walk,’ ‘see,’ etc. 
At least a hundred, and preferably more, such words should be 
used. We may find anywhere from nearly one hundred percent to 
nearly zero that are common to the two lists. The proportion is 
then a measure of relatedness. We are left with the same dilemma 
as before: What percentage of common vocabulary is necessary to 
indicate that two speech forms are varieties of the same language? 
In addition there is a new problem: How are we to determine 
whether two words in two speech forms are the same? 

The test of common elements need not be restricted to vocabu¬ 
lary. We can make similar examination of structural elements of 
any kind. Presumably, every language has a grammar, and that 
grammar must be shared among all forms of the language. But we 
know that there are grammatical differences between dialects, so 
that we cannot set up structural identity as our criterion. What 
degree of structural similarity is to be expected within a language? 
What within a dialect? 

23.6 Both these criteria have a common failing. Consider an 
area where a number of small communities, say villages, can be 
selected for examination. We will choose a number of such villages 
lying more or less in a line, each a short distance from the next. 
It could well be that the dialect A of the first village would be 
mutually intelligible, or have a high percentage of common ele¬ 
ments, with B of the second village. Similarly B and C might be 
indicated as being closely related, and so on through each suc¬ 
cessive pair until we reach Y and Z. But if we now compare 
directly the dialects A and Z, spoken some distance apart, we may 
find a degree of mutual intelligibility approaching zero and a com¬ 
paratively small percentage of common elements. Obviously, we 
must conclude that A and Z are to be included in two separate 
languages. But where are we to draw the line? Each successive 

336 Language Classification 

pair of dialects examined was found to be closely related. We seem 
to be in the dilemma of drawing a line between A and Z without 
drawing it between any two successive dialects. 

This is not merely a theoretical consideration. There are many 
places in the world where just this sort of situation is closely 
approximated. The vast area of Africa in which Bantu languages 
are spoken is a case in point. Within this area there are many 
places where such gradual transitions connect mutually unintel¬ 
ligible speech forms. Not every local vernacular is so joined to all 
its neighbors, of course, but the phenomenon is sufficiently common 
to make a satisfactory division of the area into languages essen¬ 
tially impossible. If the several classifications of the Bantu lan¬ 
guages which have been attempted are compared, it will be found 
that there are only a few places in which any measure of agreement 
is reached. While the inadequacy of the data has contributed to 
this uncertainty, the major cause of this disagreement among 
Bantu specialists is the inconclusive nature of the facts, particu¬ 
larly the considerable number of cases of intergradation. 

23.6 Another possible method would be to look for conspic¬ 
uous bundles of isoglosses, or if detailed mapping of linguistic fea¬ 
tures is not available (the more usual situation), to look for the 
more abrupt transitions. This criterion differs from the last two in 
that we have shifted our attention from the comparison of speech 
forms to the boundaries between them. In some cases this can be of 
great help. If we are faced with a problem such as that sketched 
in 23.5, we may resolve it this way: Since A and Z are evidently 
distinct languages, there must be a division between them some¬ 
where. We cannot place this dividing line in such a way as not to 
separate closely related speech forms, but we can elect to place 
it at that place between A and Z where there is the most rapid 
transition. This would be either between those local dialects which, 
though mutually intelligible, were least so; or, if information is 
available in this form, in that place where there is the most con¬ 
spicuous bundle of isoglosses. 

23.7 A rather neat refinement of the method of common ele¬ 
ments is available. .4s was pointed out in Chapter 16, it is possible to 
make a phonemic analysis based on a single dialect of English, or to 
evolve an overall pattern based on a wide range of dialects. In the 
case of English phonology, such an overall pattern fits the facts of 

Language Classification 337 

any one of a large number of dialects with quite acceptable pre¬ 
cision. We might say, then, that any group of dialects is a single 
language if it is possible to describe an acceptable overall pattern 
of some aspect of structure, say the phonology. While this is 
probably the most satisfactory criterion of all, it is a laborious test 
to apply, and does not obviate the basic difficulty. 

In the first place, how closely must the analysis fit the facts? 
There is always a certain degree of approximation in any descrip¬ 
tive statement. Any statement of overall pattern will necessarily 
be a less adequate treatment of any given dialect than a statement 
based exclusively on that dialect and done with equal care and 
methodological adequacy. How loosely can an overall pattern fit, 
and still justify treating the total base as a single language? 

For the descriptive linguist, such a criterion is particularly ad¬ 
vantageous, since his chief interest in delimiting languages is in 
defining the systems with which he must be concerned. In effect, 
this definition of a language would be statable in a rough form as: 
“A language is any form of speech of which a workable description 
can be made.” 

23.8 All of these criteria, and any others which can be pro¬ 
posed, inevitably fail, if it is expected that they can give a con¬ 
clusive answer to the problem of defining a language or of ranking 
speech forms as languages, dialects, or other categories. Ultimately, 
the decision must rest on the judgment of a linguist, though all 
these methods of measuring relationship can be of immense value 
in providing the basic data on which the decision must be based. 

With written languages, the difficulties are much less acute. 
These tend to be relatively uniform over wide areas, and the 
transition from one to another is generally abrupt with little inter¬ 
gradation. The differences between similar written languages may 
vary from rather minute to quite fundamental ones, so that the 
ranking problem is not entirely absent. Nevertheless, there are 
generally a number of discrete units from which to begin in 
classifying written languages. 

In the popular understanding, it is assumed that any language 
will be written everywhere the same, dialectal differences being 
restricted to speech. This can be inverted to give a general defini¬ 
tion of a language as a form of communication which is written 
everywhere the same, and of a dialect as a variety of a language 

338 Language Classification 

which is spoken differently from others. Innumerable difficulties 
can be seen in such a definition, some of which have been pointed 
out in Chapter 22. Nevertheless, it is some such idea that the 
layman commonly has in mind when he asks the linguist, “How 
many languages are there?” 

23.9 If we leave the search for a criterion to distinguish lan¬ 
guage from dialect and approach the wider problem of classifica¬ 
tion of languages to show wider relationships, we will find that 
two of these methods can be sharpened to become rather precise 

The first of these is common vocabulary. Consider two lan¬ 
guages, sufficiently different that the question of considering them 
as a single language does not occur. We have at hand what seem 
to be reliable vocabularies of each. These we can compare in such 
a way as to produce an estimate of the proportion of the vocabulary 
which seems to be similar. By this we need not mean anything very 
precise at first. If two words look more or less similar in form and 
have meanings which are suggestive of a relationship, we will count 
them. From this we will emerge with a percentage figure expressing 
the common vocabulary. 

The question now arises, what, if anything, is the significance 
of this percentage? We must ask this question, because there is 
always a possibility that words can be similar by mere chance. 
We must satisfy ourselves that this does not offer a plausible 
hypothesis to explain the percentage of common vocabulary which 
we have found. To do this, we must estimate what percentage it is 
reasonable to expect by chance. We can do this by selecting pairs 
of languages which we have every reason to believe are not related 
in any way, perhaps one American Indian language and one lan¬ 
guage from Central Africa. Any similarities between such a pair 
may be considered as due to chance. One investigator found that 
chance would produce about 4 percent of words “similar” in the 
degree he was considering. Suppose we have found 20 percent of 
similarities in our test case. Is this significantly more than 4 per¬ 
cent? Only a statistical measure that takes account of the size of 
the samples will tell, but if we have examined a few hundred pairs 
of words, such a measure will show that there is a high probability 
that the difference is significant. 

23.10 The results of this calculation are merely a statement 

Language Classification 339 

that it is highly probable that the percentage of common words 
which we found in the two languages A and B is due to something 
other than chance. The question is what? There are two very 
likely factors: borrowing (either from each other or from a com¬ 
mon source), and/or descent from a common ancestor by the 
familiar processes of linguistic change. The latter would indicate 
that the two languages are related in the sense with which linguists 
are most often concerned. Our statistical investigations have 
accordingly demonstrated that relationship is a possibility. But 
they cannot establish it as a fact. To do so, we must demonstrate 
that some at least of the similar words are actually cognate, by 
which we mean, descended from some common ancestral language. 

23.11 Such a demonstration depends on our observation that 
certain kinds of linguistic change (phonetic and phonemic) are reg¬ 
ular. This regularity of change may, under favorable circum¬ 
stances, make it possible to recognize those items which have been 
produced by this kind of change. If change is regular in both lines 
of descent, it follows that there should be regular correspondence 
between forms in one language and cognate forms in the other. 
That is, suppose the ancestral language has a phoneme /X/ which 
by regular change becomes /Y/ in language A, and by a different 
regular change becomes /Z/ in language B. We might then expect 
many words with /Y/ in A to correspond with words with /Z/ 
in language B. 

Conversely, suppose we observe that many words in language A 
containing the phoneme /P/ seem similar to words in language B 
containing /Q/. We may conclude that this is presumptive evi¬ 
dence that there was in the common ancestor a sound (perhaps a 
phoneme or perhaps an allophone) which by separate changes 
became /P/ in A and /Q/ in B. We do not know what that sound 
was, but we may designate it by an arbitrary symbol, say [*R]. 

We may find a great number of such correspondences. If this is 
the case, we can postulate a considerable number of allophones for 
the language, and so reconstruct the parent words for many of our 
corresponding vocabulary items. From these we can make a 
phonemic analysis, grouping the postulated allophones into recon¬ 
structed phonemes. 

23.12 In a sense, what we have just described is the phonemic 
analysis of a dead and probably unknown language from the 

340 Language Classification 

evidence preserved in its descendants. But in another, perhaps 
more accurate sense, it is the discovery and description of an over¬ 
all pattern of structure covering certain parts of the vocabulary of 
the two languages. We can make no claim that our analysis is an 
overall pattern for the two languages as wholes, since it is based on 
detectable correspondences in selected portions of each language. 
Those words in which correspondences cannot be found are either 
assumed to be affected in some way by non-regular linguistic 
change (analogic change or borrowing, for example), or to ex¬ 
emplify regular phonetic or phonemic changes which we have not 
yet been able to detect. 

23.13 As an example of the method, we may consider the re¬ 
construction of some Proto-Central-Algonquian consonants. This 
reconstruction is based primarily on the comparison of Fox, Cree, 
Menoraini, and Ojibwa. These four languages are rather closely 
related, and show quite evident similarities in many sets of words. 
Consider the following example: 

Fox Cree Menomini Ojibwa 

pematesiwa pimatisiw peraatesew pimatisi ‘he lives' 

pSsiwa posiw posew p5si ‘he embarks’ 

ncwSpam&wa niwSpam&w newapamaw niwapama ‘I look at him’ 
wfipanwi wfipan wapan wapan ‘it dawned’ 

niyawi nlyaw ngyaw nlyaw ‘my body’ 

kenosiwa kinosiw kenbsew kinosi ‘he is long’ 

The Fox words are generally longer than the others, whereas 
the Ojibwa are often shorter. We might assume that the develop¬ 
ment of Fox involved adding final vowels, or conversely, that the 
others have lost them. The first alternative seems less likely, since 
it would be necessary to account for the addition of /i/, /wi/, or 
/a/ in different contexts. It is easier to assume that these are 
original, and that any final vowel is dropped. There is much addi¬ 
tional evidence that the longer forms are older, and our recon¬ 
struction will be based on this hypothesis. All the sets of presumed 
cognates that have /p/ in any one language have /p/ in all. We 
assume that this is inherited, and so set up a Proto-Central-Al¬ 
gonquian [*p], which we may define as representing the cor¬ 
respondence /p p p p/. Similarly, we may set up [*ra], [*t], [*s], 

Language Classification 341 

[>1 [*nj, [*yj, and [*k], representing /m m m m/, etc. (The 
second /w/ in Fox /wSpanwi/, which has no correspondents in 
the other languages, is taken care of by our statements about loss 
of word endings.) The six sets of words may be represented by 
[“pematesiwa *posiwa ‘newapamawa *wapamvi *nlyawi *keno- 
siwa]. These are merely formulae from which we can predict the 
actual forms in the four languages under discussion. Only second¬ 
arily are they guesses as to the actual forms in Proto-Central- 
Angonquian as the ancestral language of the group. 

23.14 A wider sample of the languages will necessitate some 
modifications of our first conclusions. Consider the following addi¬ 
tional sets of words, where attention is directed to the phonemes 
transcribed in boldface: 










‘ my sister-in-law ’ 










‘he breathes’ 

In these sets we find two new correspondences which can be dis¬ 
tinguished from [*n] (/n n n n/) only in Cree: these are /ntnn/ 
and /n ynn/. These are no chance phenomena, since we can find 
a number of sets of words for each of them which are quite regular 
in every other respect. These two must be added to our system of 
Proto-Central-Algonquian reconstructions, and for this purpose 
the symbols [*0] and [*1J are customarily selected. Thus we 
may reconstruct: [*aOemwa *nI0emwa *elenyiwa *lchl6wa]. One 
reason for choosing the symbols £*0[] and [*1] is that Arapaho has 
these correspondences as /0/ and /!/, and that many other lan¬ 
guages have /l/ for [*1]. Illinois is derived through French from 
the Algonquian word ‘man' in one such language. However, we 
must not assume that we actually know precisely how either of 
these was pronounced in the ancestral language, nor indeed dare 
we be dogmatic even about [*p3 or C*m]. 

In all of these reconstructions, and most noticeably the last, we 
have had to make use of some correspondences which have not 
been mentioned: those between the vowels, and /n hy hn ss/ which 
is noted by [*hl]. No single correspondence can be considered as 
established until enough have been found to enable the worker to 

342 Language Classification 

reconstruct a number of words or morphemes. The consonant 
correspondences are of very little value unless we can also account 
for the vowels. This is another instance of the principle which has 
been mentioned in other contexts: We cannot establish the pho¬ 
nemes individually without regard to their place in the system as a 
whole. We cannot consider a morpheme as established unless we 
can account for the rest of the utterance in which we find it. 
Languages are integrated systems, and we cannot treat them 

23.16 One more set of words should be commented on: 

6sani ohtfiwiya Ohnan Sssan ‘his father’ 

With the exception of the Cree form, this is reconstructed as 
[*ohOali]. Actually, the first part of the Cree word fits, since from 
this reconstruction we would expect a form like /*ohtan/. We can 
only assume that some sort of analogic change has taken place in 
this particular word at some point in its history in Cree. No 
comparative reconstruction can be expected to fit the entire vo¬ 
cabulary in any one of the languages compared. There will be 
sets in which a partial analogic reformation can be easily detected 
and a reconstruction made on the basis of the evidence available. 
There will also be instances in which no reconstruction can be 
made at all, since there may be no certain evidence to identify any 
one word as inherited from the ancestral language. Reconstructions 
never profess to comprehend the entire vocabulary and structure 
of the languages on which they are based, but only parts which 
can be proven to be cognate. 

The procedure which has just been described is one of the tools 
of comparative linguistics. This discipline is coordinate with de¬ 
scriptive linguistics as a second major branch of language science. 
The methods of the two are basically very similar. They differ 
largely in the selection of the data with which they are concerned. 
Rigorous methods of language study were first developed for cer¬ 
tain comparative problems. Descriptive linguistics therefore owes 
many features of its methodology to comparative linguistics. In 
return, comparative work, to attain the maximum result from its 
method, must depend on descriptive linguistics for its data. For 
example, the reconstruction of Proto-Central-Algonquian just 

Language Classification 343 

mentioned is possible only because adequate analyses had been 
made of the four languages concerned. 

23.16 The methods of comparative linguistics enable us to 
demonstrate that two languages are related. It remains to deter¬ 
mine the bearing of this relationship on the classification. 

Total loss of morphemes is an inevitable feature of linguistic 
change. There is evidence that this loss proceeds at about the same 
average rate for all languages. A number of calculations have in¬ 
dicated that this rate is about 19 percent per thousand years for 
the most basic vocabulary. Suppose two languages separate com¬ 
pletely so that their subsequent histories are independent, and that 
in the thousand years subsequent to separation each language 
loses 19 percent of the morpheme stock they had in common when 
they were a single language. Each language retains 81 percent of 
the original stock. Suppose we consider an original sample stock 
of 200 morphemes. Language A will retain about 162 of them, as 
will language B. But there is no reason to expect that the two 
languages will necessarily lose the same items. The most probable 
outcome is that language B will retain 81 percent of the 162 which 
language A retains, as well as 81 percent of the 38 which language 
A losses. This means that A and B can be expected to have about 
132, or 66 percent, of this basic stock in common. 

Such a calculation can be reversed. If 66 percent of the basic 
morpheme stock seems to be cognate in two languages, we may 
assume that they have been separate for 1000 years. If 44 percent 
is cognate, 2000 years is the most probable period of separation. 

This method, known as glottochronology, still in the early stages 
of development, promises to provide a useful basis for interpreting 
the degree of language relationship. It also provides a means of 
dating certain events in prehistory. Such dates, like Carbon-14 
dates, are statistical. They provide only an estimate of the most 
probable date, together with some estimate of the probability of 
any given deviation from such a date. 

23.17 The methods of comparative linguistics are laborious 
and depend on good descriptive materials being available for the 
languages concerned. They have not as yet been successfully ap¬ 
plied to more than a few language groups. The methods of glot¬ 
tochronology, if they are to be rigorously applied, must follow 
comparative analysis, since they depend on accurate identification 

344 Language Classification 

of cognates. This is possible only when a systematic analysis of 
correspondences has been established. 

For large areas, languages can therefore be classified only on 
the basis of direct and uncontrolled comparisons of vocabulary 
often poorly recorded. This means that our present ideas of lan¬ 
guage classification must be considered as nothing more than 
tentative. In general, it may be said that the more refined methods 
are of greatest need when more remote relationships are concerned. 
Closely related languages are usually clearly marked by evident 
similarities, so that intuitive conclusions based on examination of 
whatever data may be available arc often quite reliable. Super¬ 
ficial resemblances between more distantly related languages may 
be absent, but sufficient evidence of a less apparent kind can often 
be found by comparative study. However, there is a degree of rela¬ 
tionship at which evidence is so scanty that it cannot readily be 
separated from the chance resemblances which can occur between 
any two languages. When we say that two languages are not 
related, we merely mean that the relationship, if any, is so remote 
that the most powerful methods yet applied to the data cannot 
detect it. We cannot prove conclusively that any two languages 
are not related, only that they do not seem to be related at a 
certain level. 

23.18 Africa will serve as a case study for some of the problems 
of language classification. The Bantu languages are very closely 
related and show numerous marked common characteristics in 
grammar and vocabulary. In addition, they occupy a generally 
compact area from the equator southward, almost to the Cape. 
Their close relationship was noted at a very early stage in the 
study of African languages and is universally accepted. Since the 
group is large, it was assumed to represent a language family (the 
largest taxonomic group usually recognized). Similarly, another 
group of languages in the northern and northeastern parts of the 
continent were recognized as related, though there was no agree¬ 
ment on the precise limits of the group. This was designated as the 
Ilamitic family. Between the two lay a large number of languages 
with no evident relationship to either, nor much with each other. 
These were designated as Sudanic. Some workers recognized that 
this was merely a convenient term to cover an assortment of lan¬ 
guages whose affinities were not understood (a necessary and useful 

Language Classification 345 

device at a certain stage of knowledge). But others misinterpreted 
the Sudanic group as in some way coordinate with the Hamitic 
and Bantu families. The result has been serious misunderstanding. 

These three African language “families” were commonly defined 
typologically. Thus Bantu languages were described as having a 
well developed system of noun classes with prefixed class markers 
and a highly developed system of concord. Hamitic languages were 
defined as having, among other features, clearly developed gender. 
Sudanic languages were alleged to be monosyllabic and tonal. 
Actually, none of these characterize any of these groups. There are 
Bantu dialects in which noun classes and the concord system are 
nearly or wholly lacking. Many Sudanic languages (e.g., Bariba, 
cf. 11.8) have equally complex systems of classes and concord. A 
typological classification may be useful for certain purposes, but 
such a classification as this, which was generally interpreted as 
genetic, though typologically defined, is misleading in the extreme. 

23.19 In recent years great advances have been made in our 
knowledge of African languages. Various elements in a better 
classification have gradually emerged. First various groups of 
Sudanic languages were sorted out and shown to be closely related. 
Then various of these groups were shown to have common descent, 
and the various resemblances between the largest group of these 
and Bantu were pointed out. In the meantime, evidence for the 
falsity of the old typological criteria was accumulating. For ex¬ 
ample, a close study of the root morphemes of several West African 
Sudanic languages was published which proved conclusively that 
some, at least, were far from monosyllabic. 

The result was that in 1949 and 1950 a new scheme for the 
classification of the African languages was advanced and is rapidly 
winning general acceptance. This unites the largest part of the old 
Hamitic Family with the Semitic languages (centering in Asia) 
and one group of the old Sudanic group into the Afro-Asiatic 
Family. It unites the greater number of the Sudanic languages of 
West Africa with Bantu languages into the Niger-Congo Family, 
and it sets up thirteen smaller families to contain the remaining 
Sudanic languages. These were not claimed to be necessarily inde¬ 
pendent, but only to be not as yet demonstrated to be related. 
Since then, four of the thirteen have been united by the discovery 
of evidence of their common descent and may be known as the 

346 Language Classification 

Chari-Nile Family. The new classification cannot be considered 
as definitive in all details, but it does at least clear the air of the 
old typological confusions and provide a basis for progress. It 
would seem to be essentially correct in its main outlines. 

23.20 With regard to other parts of the world, our knowledge 
of linguistic classification is in various of the stages through which 
African linguistics has passed. In South America no real classifica¬ 
tion has as yet emerged. In New Guinea languages are classified 
into two groups: one, Malayo-Polynesian, is like Bantu in being a 
clearly marked genetic unity; the other, Papuan, is like Sudanic in 
being only a taxonomic wastebasket out of which we must some 
day sort groups showing relationship. In some areas our knowledge 
has passed beyond that now reached in Africa. In Europe, for 
example, most of the problems of language classification have been 
met, and a great deal of detailed comparative work has been done. 

23.21 The second reason given for our ignorance of the number 
of languages in the world is the inadequacy of our knowledge. In 
large areas we have only the most rudimentary information. There 
has recently appeared a volume purporting to list the languages 
and dialects of West Africa. The auspices under which it appeared 
are such as to guarantee high quality and near exhaustiveness 
within the limits of total scientific knowledge to date. We may 
therefore take West Africa as a type example to illustrate the prob¬ 
lems. There are other areas in which our knowledge is much more 
complete, e.g., Europe, and some in which it is much less satis¬ 
factory, e.g., New Guinea. 

Of the considerable number of languages listed, at least a third 
are entered with some reservations. For many, all that can be said 
is that some traveler reported a language of that name in a certain 
region. For many of these, we have no way of judging the actual 
status. Some may actually be languages. Others are probably 
poorly marked dialects of some neighboring language listed else¬ 
where. In a few cases, the reported “language” may actually not 
designate a speech form at all. Many of these reports are made by 
people who know very little about linguistics or the languages of 
the area. Moreover, if some languages are known only from such 
casual reports, what assurance have we that there are not many 
others which have not been reported at all? 

Sometimes what is reported as a single language is later found 

Language Classification 347 

to be a group of two or more speech forms, perhaps not even 
closely related. An extreme case of this sort is the “ Mimi” language 
of French Equatorial Africa. This is known from two brief accounts 
by travelers. But the two are certainly not based on the same 
language, and in fact seem to treat two languages of different 
language families. How many other “languages” are actually 
similar groups? 

23.22 Of course the descriptive work that can be done on any 
language is enormous. No language can be said to be fully de¬ 
scribed. It would be utopian to wish for anything approaching 
complete description for all the languages of the world. To enable 
us to compile an adequate treatment of the languages of the world 
in handbook form a minimum objective might be as follows: 

For every language and major dialect, there would be needed at 
least the following: a summary of the phonology and morphology; 
a vocabulary of, say, several thousand words; a small body of 
recorded texts together with a translation into some better-known 
language; a statement of where and by whom the language is 
spoken, and the name by which it is called by its speakers and the 
neighboring peoples. With this much information, a definitive 
listing and classification of languages could be made. 

Beyond this, there is frequently need of a much more com¬ 
prehensive record. The socially more important languages, and a 
few sample languages from each group, whether socially important 
or not, should be described much more adequately. Incredibly few 
languages have, for example, an adequate dictionary, that is, one 
in which the vocabulary entries run to tens of thousands and in 
which the definitions are not merely one word translations, but 
include citations of usages and discussion of the range of meanings. 
Many more such dictionaries are needed. Still fewer languages have 
adequate grammars. Too many of the grammars we have are 
arbitrarily fitted into a mold that is irrelevant or actually con¬ 
trary to the structure of the language. Serious treatments of syntax 
are exceedingly rare, and treatments of intonation or similar 
phenomena are almost non-existent. The basic task of descriptive 
linguistics, that of describing languages, is very far from ade¬ 
quately discharged. 

23.23 If all the world’s supply of descriptive linguists could 
be put on the job of making minimally adequate descriptions of 

348 Language Classification 

speech forms under a well financed and carefully directed pro¬ 
gram, it might be that we could meet the first goal within a decade. 
But apart from the obvious financial limitations, this is not at all 
feasible. There are other important activities to occupy the time 
of linguists: perpetuating the profession by training replacements, 
making their contributions to “liberal” education and the training 
of specialists in other fields, conducting fundamental research for 
the advancement of linguistic theory and method, working on the 
application of their findings to various practical problems like 
the teaching of languages, and carrying on a program of public 
relations to acquaint the public with the work which is done and to 
recruit new personnel for the field. There simply are not enough 
competent descriptive linguists, nor have they finances and facil¬ 
ities with which to work. 

Yet there is a certain urgency in the needs. From a scientific 
point of view there is an urgency because the data is being lost. 
Languages are changing or disappearing. Some of them might make 
important contributions to a basis for a more comprehensive 
theory. There is also a practical urgency. Almost daily, decisions 
are being made which should presuppose adequate knowledge of 
the languages of the peoples concerned. Orthographies are being 
established and new written languages launched. Vernacular edu¬ 
cation is being pushed, sometimes with no clear knowledge of the 
languages concerned or the special problems they present. Efforts 
are being made to extend government or trade languages at the 
expense of smaller speech forms; sometimes more information 
would make possible a transition with less disruption. There is in¬ 
creased need for competence in less known but often widely used 
languages; often our knowiedge of their structure is not adequate 
for effective teaching. 

23.24 All this points to the need of more descriptive linguists. 
It has recently been estimated that there are only about three 
hundred in the United States. A decade ago the number was much 
less, and presumably increase will continue. Nevertheless, there 
is no prospect that there will be enough professional linguists to 
do all the field work that needs to be done. 

It would seem therefore that there is an excellent opportunity 
for amateurs in this field. In the past, a great deal of linguistic 
w'ork was done by such people, missionaries, colonial officials, 

Language Classification 349 

business men, and others. Their work was not always all that might 
be desired, but much of it was done before the development of 
modern methods, that is, at a time when a professional could have 
done little better. In the present time, one of the most urgent needs 
of the science is to make its basic concepts and methods known, so 
that such amateurs can produce results that will be acceptable by 
modern standards. This should be quite practical, since many of 
these people have some direct interest in learning the languages 
concerned, and can directly profit by many of the concepts and 
techniques in their own language learning. 

One of the most monumental undertakings in the field of lan¬ 
guage study was the New English Dictionary. For this, several 
million quotations from an exceedingly wide range of English 
literature were assembled by over 1300 readers, most of them 
volunteers. Although no such organized cooperative effort has ever, 
to my knowledge, been carried out within the field of descriptive 
linguistics, and few in the closely allied fields of geographical or 
social linguistics, it would seem quite possible to do so. The ex¬ 
amples of such sciences as astronomy, with a well organized 
program of comet and meteor observation largely by volunteer 
hobbyists, and meteorology, with heavy dependence on volunteer 
weather recorders, would suggest that if such forces could be 
organized they might prove very valuable. But apart from such 
large-scale undertakings, there are certainly many ways in which 
individual workers can assemble important data as a hobby or 
incidentally to other occupations. For greatest effectiveness, this 
work would require close cooperation between professionals and 
amateurs. Given such cooperation, we may expect rapid advance 
in our knowledge of the languages of the world in years to come. 





Some Languages and 
Language Families 

24.1 The largest and most important language family, from the 
point of view of both the social importance of the major languages 
in the group and their interest to linguists, is the Indo-European. 
The comparative method was very early applied to the study of 
this family, and more comparative work has been done on Indo- 
European than on all other groups together. As a result, many 
features of the family are well known. There is relatively little 
debate about the limits of the family or the major groups into 
which it can be divided; but there is still difference of opinion as to 
the details of interrelationship of the groups. This is largely be¬ 
cause the classical methods of comparative linguistics were not 
adequate to give a decisive answer in this matter. It remains to be 
seen whether newer methods will succeed. The next several sections 
will discuss very briefly the major branches of the Indo-European 

24.2 The Germanic languages consist of three groups of im¬ 
portant languages: English-Frisian, Dutch-German, and Scan¬ 
dinavian. The first includes only English, with more speakers than 
any other language of the present day, and Frisian, spoken by a 
relatively small population along the coast in the Netherlands and 


Some Languages and Language Families 351 

Germany. On the continent of Europe the Dutch-German language 
area supports three well-known written languages, German, Dutch, 
and Flemish. (See 22.10.) The first two, especially, have been 
carried abroad and are spoken in many parts of the world. Afri¬ 
kaans, one of the two official languages of the Union of South 
Africa, is a development from Dutch. Yiddish is basically a 
German dialect with a written language using the Hebrew al¬ 
phabet. Continental Scandinavia supports four written languages: 
Danish in Denmark, Swedish in Sweden and Finland, Rilcsmdl 
and Landsm&l, two competing writing conventions in Norway. 
Icelandic is also of the Scandinavian group. The oldest extensive 
documents in any Germanic language are the Gothic Scriptures 
translated by Wulfila in the fourth century, of which, unfortu¬ 
nately, only a part has been preserved. The most ancient Scan¬ 
dinavian records of any length are in Old Icelandic from the 
twelfth century. The documents from the Dutch-German area 
before the rise of the modern standard languages represent a 
number of dialects, of which the most commonly mentioned in¬ 
clude Old Saxon and Old High German. Old English is the best 
designation for the Germanic dialects of Britain before the Norman 
conquest, though the less suitable Anglo-Saxon is often used. 

24.3 Of the Celtic languages, formerly much more widespread, 
only four retain any vitality. Breton in the extreme northwest of 
France competes with French and is slowly losing ground. Welsh, 
Irish, and Scots Gaelic resist submersion in English only by isola¬ 
tion and local nationalism. 

24.4 The Romance languages contain five very important writ¬ 
ten languages: Portuguese is the language of Portugal and Brazil 
and the official language of the Portuguese Empire in Africa and 
Asia. Spanish is the language of the larger part of Spain and of 
most of Latin America other than Brazil. French is the official 
language of France and the French overseas territories, and one of 
the official languages in Belgium, Belgian Congo, Switzerland, and 
Canada. Italian is the official language of Italy, and the spoken 
language of many Italian emigrants. Roumanian is the official 
language of Roumania. In the Romance area, language and 
political boundaries seldom coincide with any precision. Within 
Spain the well-marked Catalan dialect area is commonly recog¬ 
nized as a separate language, and the Galician dialects are more 

352 Some Languages and Language Families 

closely related to Portuguese than to Spanish. In the south of 
France the local dialects are known collectively as Provencal; but 
for the political accident, they might have provided the base for 
another important written language. Sardinian is quite distinct 
from Italian. In the Alpine regions of northeastern Italy and 
adjoining Switzerland is a group of closely related dialects known 
to linguists as Rhxto-Romanic. One of these, known as Rotnansch, 
has developed a written form and become (with German, French, 
and Italian) one of the official languages of the Swiss Republic. 
In some areas outside of Europe, local vernaculars derived from 
Romance languages have developed. Probably the most clearly 
marked and most important socially is Haitian Creole. 

Classical Latin, as known from its extensive literature, stands 
very close to the common ancestor of all the modem Romance 
languages. The latter must, of course, have been a spoken lan¬ 
guage; this is sometimes referred to as Vulgar Latin, but this 
term is often used in a very inexact way. In ancient times there 
were a number of other Indo-European languages spoken in Italy. 
Not all were closely related and many arc so poorly documented 
that no clear evidence of their relationships is available. Those 
related to Latin are designated as the Italic branch. Oscan, Um¬ 
brian, and Venetic belong with Latin in this group. Etruscan, the 
language of one of the most important peoples in ancient Italy, is 
not Indo-European, and no relationship has been established. 

24.6 A large part of Eastern Europe is occupied by speakers of 
Slavic languages. About a half of this population use Russian, 
originally the language of the region centering on Moscow but now 
spoken across northern Asia, in which area it is supplanting many 
of the indigenous languages. It is also widely used as a second 
language in portions of the U.S.S.R. in which other languages are 
dominant and in countries of the Soviet sphere. This gives it a 
position as one of the leading languages of the world, probably 
second only to English in social and political significance. East 
and south of the Russian area within the Soviet Union are the 
slightly different Byelorussian and Ukrainian languages. Three 
other Slavic languages served as the nuclei around which inde¬ 
pendent states were organized following World War I. These are 
Polish, Czech and Slovak (dialects of one language), and Serbo- 
Croatian (with its two written languages, Serbian and Croatian). 

Some Languages and Language Families 353 

354 Some Languages and Language Families 

Bulgarian has served in the same way for a somewhat longer time. 
Old Church Slavonic is the first written language of the Slavic 
group, dating from the ninth century. It is still used as a liturgical 
language in some of the Orthodox Churches. 

24.6 Lithuanian and iMluian arc the only two languages of 
any social importance in the Baltic branch. Albanian , with no 
known close relationship comprises a branch of its own. Armenian, 
spoken in the southern Caucasus and in scattered communities in 
the Near East and elsewhere, is another isolated Indo-European 

Modem Greek, together with the various ancient and medieval 
forms of Greek, comprises a branch. Ancient Greek is a complex of 
dialects representing successive waves of Indo-European-speaking 
peoples spreading into the Aegean area. Recent decipherment has 
extended the known history of Greek to over three millennia. 
Within such a time-span there must, of course, be considerable 
change, so that it is important to maintain a careful distinction 
between the different stages. Unfortunately it is common practice 
to speak merely of “Greek” without recognizing these differences 
or indicating the form of Greek which is meant. 

24.7 The Iranian branch contains four important spoken lan¬ 
guages or perhaps groups of closely related languages. In the 
mountains of eastern Turkey, Iraq, and western Iran is Kurdish. 
A large part of Iran uses Persian, and Persian is also an important 
second language among Muslims in India and Pakistan. Part of 
Afghanistan and adjacent areas in Pakistan use Pashto or Afghan. 
Balochi is the main language of Baluchistan in Pakistan. A large 
number of older Iranian languages have left important literary 
remains. The oldest is Old Persian, known from the sixth to the 
third century b.c. Avestan is the language of the Zoroastrian 
scriptures. Pahlavi was used in the Persian Empire of post- 
Christian times. 

24.8 The Indie branch includes most of the languages of 
northern India and Pakistan. There are a very considerable num¬ 
ber of languages in the branch, and several of them are spoken by 
large populations. A classification based on the spoken dialects 
does not accord with the usual understanding of the people in 
India and Pakistan, since Hindi and Urdu are two literary lan¬ 
guages which arc used in association with widely divergent spoken 

Some Languages and Language Families 355 

forms. Hindi is now the official language of the Republic of India, 
and Urdu is official in Pakistan. Bengali, Assamese, Oriya, Marathi, 
Gujerali, Sindhi, Panjabi, Kashmiri, and Nepali are among the 
best-known of the group. Sinhalese, the chief language of Ceylon, 
though far removed from the main area, is also Indie. 

The Indie branch has a long literary history. A large part of this 
literature is in Sanskrit, still widely used as a literary and liturgical 
language in India. Sanskrit is of great interest to linguists because 
of the high development of descriptive linguistic technique cul¬ 
minating in the work of Panini in the fourth century b.c., and 
because of the stimulus which the introduction of Sanskrit to 
Western scholarship gave to the development of modern linguistic 
science. The Vedas, in a language related to classical Sanskrit, are 
the oldest documents in any Indo-European language, though we 
have them only in much later copies. The other older Indie 
languages are known collectively as Prakrits. 

The Indie and Iranian languages are sometimes classed together 
as Indo-Iranian. 

24.9 There are numerous extinct Indo-European languages of 
which we have some records. These range from rather extensive 
collections of documents to a few short inscriptions. The better 
attested ones can be classified with some certainty. The correct 
place for some of the others is still in question after exhaustive 
analysis of all the available data. Among the best known are those 
in the Tocharian branch, two languages spoken in Central Asia 
from the seventh to tenth centuries. These are of interest to com¬ 
parative Indo-European studies because they add a whole new 
branch to the data. The other Indo-European languages which 
cannot clearly be assigned to any of the branches enumerated 
above are largely known from the Mediterranean area. Among 
those most commonly mentioned are Illyrian and Phrygian. 

24.10 Hittite is known from a large number of inscriptions and 
tablets beginning about a millennium and a half b.c. Together with 
a number of other ancient languages also from Asia Minor, 
Hittite is classed as Anatolian. These languages show evident 
relationship to Indo-European, but the exact nature of this con¬ 
nection is not agreed upon. Some have considered Anatolian as 
another branch of Indo-European. The majority consider that 
Anatolian separated from Indo-European before the latter broke 

356 Some Languages and Language Families 

up into the present branches, so that Anatolian and Indo-European 
together constitute a larger group sometimes referred to as Indo- 

24.11 The Finno-Ugric family includes three socially important 
languages of Europe: Finnish, Estonian, and Hungarian. In addi¬ 
tion, there are various others scattered over northern Europe and 
Asia. These include Lappish , Mordvin, Cheremiss, Votyak, and 
the Samoycd languages. 

24.12 The Altaic family consists of three branches. The Turkic 
languages include Turkish; Azerbaijani spoken in northwest Iran 
and the Caucasus; and a number of languages or dialects in central 
Asia, notably Kirghiz, Uzbcg, Turkoman, and Kazak. The Mongol 
branch includes several closely related languages usually lumped 
under the label Mongolian. The third branch consists of Manchu 
and Tungus spoken to the east of Mongolia. 

24.13 The Caucasus is an area of extreme linguistic diversity. 
In addition to Armenian, Azerbaijani, Iranian languages, Russian, 
and Ukrainian, which have already been mentioned, there is a 
large number of languages frequently classed together as Cauca¬ 
sian. This is not entirely satisfactory, since there is no demon¬ 
strable relationship between some of these, and they should be 
classed into two families. The South Caucasian family includes 
Georgian , with an important and ancient literature, and Min- 
grelian. The North Caucasian family contains a very large number 
of languages, the best known of which are Abkhasian, Avar, 
Chechen, and Kabardian. 

24.14 In the western Pyrenees in France and Spain a small 
population speaks Basque. Its affinities are one of the recurrent 
topics of debate and speculation, and relationship to almost every 
linguistic family in the Old World has been proposed. None of the 
proposals seems capable of proof by any acceptable methodology, 
so that Basque remains an independent family. 

24.16 The Afro-Asiatic family is so-called because it is spoken 
in northern Africa and southwestern Asia. There are five branches, 
Semitic, Egyptian, Berber, Cushitic, and Chad. The last four are 
sometimes known as Hamitic, and the whole family as Hamito- 
Semitic. The term is unfortunate, however, because it implies 
that the four “ Hamitic” branches are more closely related to each 
other than to Semitic, which does not seem to be the case. 

Some Languages and Language Families 357 

24.16 The Semitic branch is the best known. Only Hebrew, 
Arabic, and some of the languages of Ethiopia are of present im¬ 
portance as spoken languages. Arabic, however, includes a number 
of very divergent dialects which, but for tradition, might be con¬ 
sidered as separate languages. Several ancient Semitic languages 
are well known and have important literature. Akkadian (also 
called Assyrian or Babylonian ) and Sumerian, which is of unknown 
relationship, but certainly not Afro-Asiatic, are the chief languages 
of the vast cuneiform literature from Mesopotamia. From these 
two languages, primarily, we have our oldest historical records. 
Aramaic and Syriac are closely related dialects. Anciently, Aramaic 
languages were dominant as vehicles of government and business 
in the Near East after the eclipse of Akkadian. One form of 
Aramaic is of present-day importance as the language of much 
Rabbinical Jewish literature, while Syriac is a liturgical language 
in some Eastern churches and the language of much early Christian 
literature. Phoenician, only slightly different from ancient Hebrew, 
was an important trade language in the Mediterranean and very 
probably the language in which the alphabet originated. A later 
form of the same language, known as Punic, was the language of 
the Carthaginian Empire. Hebrew, the language of ancient Canaan, 
is of importance chiefly as the vehicle of the larger part of the Old 
Testament Scriptures (a small part is in Aramaic). It was sup¬ 
planted before the time of Christ by Aramaic in spoken usage, but 
has continued as a liturgical and literary language and is now being 
revived as a spoken language in Israel. With this long history 
there were inevitable changes, and Modern Hebrew differs in 
various important respects from Ancient Hebrew. In the southern 
part of Arabia the vernaculars known collectively as South Arabic 
are quite different from classical Arabic or the dialects to the 
north, though classical Arabic is the written language most used 
in the area. There are three important vernacular .Semitic lan¬ 
guages in Ethiopia. Amharic, spoken in the center of the country, 
is the official language. Tigri and TigriHa are spoken in the north. 
Ge'ez (also commonly called Bthiopic) is an older language related 
to these three and is the liturgical language of the Ethiopian 

24.17 The second branch of the Afro-Asiatic languages con¬ 
sists of ancient Egyptian and its descendant, Coptic. The latter 

358 Some Languages and Language Families 

remains the liturgical language among the Coptic Christians of 
Egypt, but has been supplanted by Arabic as a vernacular. 

The Berber languages are widely scattered over North Africa 
and the Sahara. In many places they have been supplanted by 
Arabic, but are still widely used as the home language in many 
areas. Kabyle, Shilh, Zenaga, and Tuareg are among the best 

The Cushitic languages occupy most of the eastern horn of 
Africa. Somali, Galla, and Beja are spoken by the largest popula¬ 
tions and over the widest areas. 

The Chad languages include an immense number of languages 
spoken in central and northern Nigeria and around Lake Chad. 
Most of these are very poorly known, and few are spoken by 
populations of any size. Hausa, however, is one of the most im¬ 
portant languages of Africa. It is the first language of a very large 
number of people, and in addition is a language of commerce over 
a very wide area in West Africa. 

24.18 The Sudan belt of Africa contains a number of small 
language families, mostly containing only very poorly known 
languages, as well as several large groups. It is thus an area of con¬ 
siderable linguistic diversity, particularly in certain regions. 

The Songhai language spoken along the Niger River in the 
vicinity of the great bend constitutes the most important of these 
small families, since it is the language of a large population which 
has had an important part in the history of West Africa. 

24.19 Centering generally in the upper Nile Valley, but ex¬ 
tending southward to Tanganyika and westward into the Basin 
of the Chari River almost to Lake Chad, is the Chari-Nile family. 
At the center of the area is the Nilotic branch including Dinka, 
Nuer, and Shilluk. Acoli in Uganda, Masai and Nandi in Kenya 
and Tanganyika have been erroneously called “Nilo-Hamitic.” 
Actually they have no discoverable relationship to the Afro- 
Asiatic languages (which include “Hamitic"). In northern Sudan, 
languages of this group persist as remnants surrounded by Arabic. 
The best known of these is Nuba, spoken along the Nile in the 
vicinity of the Egypt-Sudan border. The languages to the west 
are very poorly known but seem to form one branch, the Central 
Sudanic. Probably Bagirmi and Moru are the most important. 

24.20 From Lake Chad northward and eastward is an area 

Some Languages and Language Families 359 

occupied by languages of the Central Saharan family. Of these, 
the only one which is well known is Kanuri, spoken in northeastern 
Nigeria and vicinity. 

24.21 The most important language family of Africa is the 
Niger-Congo. These languages are spoken in most of West Africa 
and generally in Africa south of the equator. There are a number 
of branches, but present knowledge does not provide a definitive 
classification throughout the group. 

The West Atlantic and Mandingo branches are quite clearly 
defined. Both are spoken at the western end of the area. The West 
Atlantic languages are generally found near the coast from Liberia 
to Senegal. Temne and Buiom in Sierra Leone and Wolof in 
Senegal are well known. The most important language of the 
branch is, however, Fulani, spoken by the Fula tribe in widely 
scattered areas from Senegal to Cameroons. The Mandingo lan¬ 
guages are spoken to the east of the West Atlantic group. The 
main center is in Liberia and Sierra Leone, where Kpelle, Loma, 
and Mende are spoken by large populations, and in the areas im¬ 
mediately to the north, where Malinke and Bambara are the best 

The coast from Liberia to Cameroons is largely occupied by 
languages of the Kwa branch. A number of these are of great social 
importance: Akan with its dialects Fanti and Twi in the Gold 
Coast, Baoule in Ivory Coast, Ewe in Togoland, Fon in Dahomey, 
Yoruba, I bo, and Nupt in Nigeria. Bassa and Kru in Liberia arc 
generally considered as belonging to this branch, but the evidence 
is not entirely clear. North of the area of these languages a large 
area is occupied by the Gur branch. These are by no means as well 
known as the Kwa languages, and the branch is not so well estab¬ 
lished in the classification. Mossi is probably the best known and 
of the greatest social importance. 

From the Cameroons eastward across the northern edge of the 
Congo basin are a number of Niger-Congo languages whose classi¬ 
fication is very little understood. They may be all one branch, or 
divisible into several. Zande in Belgian Congo and Sudan is cer¬ 
tainly the best known and spoken by the largest population. Sango, 
the lingua franca in much of French Equatorial Africa, also 
belongs here. 

The Central branch centers in eastern Nigeria and Cameroons. 

360 Some Languages and Language Families 

It consists of numerous sub-branches, mostly containing languages 
of which we know relatively little, spoken by small tribes. Efik 
and Tiv are exceptions in that each is used by a large population. 
One group, however, has expanded greatly, spreading southward 
and eastward over a very large territory. This is the Bantu group 
which, though it covers a larger area than all the other Niger-Congo 
languages together, is not a branch but only a subdivision of a 

There are probably more languages and dialects in the Bantu 
group than in any other group of comparable rank anywhere. 
Classification is, therefore, very difficult, and no agreement can 
be reached at many points. A number of important languages can 
be listed, however. These are of three kinds: Some are trade or 
government languages which have recently come to be very widely 
used far beyond the territory in which they were originally spoken. 
The most important by far is Swahili. This is used throughout East 
Africa and eastern Belgian Congo, and is occasionally heard in all 
the adjoining countries. In Congo three other languages have had 
similar development: Kongo in the east, Luba in the south, and 
Ngda in the north. These trade languages are used along with the 
tribal languages, but with a growing tendency to replace some of 
them. Another group of important languages are the union lan¬ 
guages. Shona, centering in Southern Rhodesia, was described 
in 22.7. Nyanja, developed on the base of a number of tribal 
dialects in southern Nyassaland, is another. Finally there are a 
number of tribal languages which, because of the size or importance 
of the tribe or other more or less fortuitous circumstances, have 
become either well known or important. Among these are Ganda 
in Uganda, Kikuyu and Kamba in Kenya, Chaga and Nyamwesi 
in Tanganyika, Rundi and Rwanda in Ruanda-Urundi, Bemba in 
Northern Rhodesia, Umbundu and Kimbundu in Angola, Hereto in 
Southwest Africa, Zulu, Xhosa, Swazi, Southern Sotho, Northern 
Sotho, Tswana, and Venda in South Africa. 

Bantu languages have a system of prefixes indicating gender 
and number. The names of the languages are sometimes given with 
these prefixes, and sometimes without. Thus Kongo is sometimes 
called kiKongo or Kikongo; Ganda is sometimes listed as Luganda, 
etc. Moreover, the tribes generally have similar names. Thus 
Luganda is spoken by the Baganda who inhabit Uganda. 

Some Languages and Language Families 361 

24.22 The spread of the Bantu languages and the peoples 
speaking them is a relatively recent historical development and 
was still actively in progress at the first European contact. Among 
the peoples displaced were speakers of the Khoisan languages. Two 
small tribes speaking languages related to this group persist in 
Tanganyika surrounded by Bantu: Sandawe and Halsa. The 
largest remnant is in South Africa, where the Bushman languages 
and Hottentot occupy a large area of sparsely populated desert 
and scrub. 

24.23 Two important Asian languages seem to be isolated 
without important relations. These are Korean and Japanese. 
Each of these must, therefore be counted .as an independent 
language family. 

24.24 The most extensive language family in eastern Asia is 
the Sino-Tibetan. It may be considered as containing two branches, 
Tibeto-Burman and Chinese. Tibetan, spoken not only in Tibet 
proper but also in many areas around the borders, and Burmese, 
the language of the dominant people in Burma, are the two largest 
languages in the Tibeto-Burman branch. The mountainous regions 
lying between these two language areas in India, Pakistan, and 
Burma are occupied by peoples speaking a very wide variety of 
mostly Tibeto-Burman languages. This area is linguistically one 
of the most diversified in the world, and it is, as yet, very difficult 
to get a clear picture of the interrelationships of the numerous 
languages and dialects. Garo, Bodo, the Naya languages, and the 
Kuki-Chin languages belong to this branch. The Karen languages 
of southern Burma are of uncertain affinities, but are usually 
treated as belonging to the Tibeto-Burma branch. 

The Chinese languages constitute the second large branch of the 
Sino-Tibetan family. Of these, the most widely used is that known 
as Mandarin. This is the language of roughly the northern half of 
China. The southeast of China is occupied by a number of lan¬ 
guages. Around the mouth of the Yangtse are the Wu dialects, of 
which Suchow is the representative most frequently cited. South¬ 
ward along the coast the linguistic diversity is considerable. These 
are commonly referred to as the Fukien dialects (local custom is to 
refer to all these languages as “dialects”). Each is known by the 
name of the city around which it is spoken, e.g., Amoy , Foochow. 
Inland there is a large area in which Hakka is spoken. South of 

362 Some Languages and Language Families 

Some Languages and Language Families 363 

Hakka is a large area in which the dialects are collectively known 
as Cantonese. In southwestern China, the local languages are 
largely non-Chinese, of various relationships, and mostly poorly 
known to linguists. Miao and Yao are most frequently mentioned. 
Mandarin Chinese is extending into the area and displacing some 
of these languages. 

24.25 The Kadai family contains a number of small languages 
of southwestern China and Hainan Island, as well as three large 
and socially significant languages or language groups. These are 
Thai or Siamese, Laolien or Lao in Indo-China, and the Shan 
languages of Burma. These languages have long been considered 
as a branch of the Sino-Tibetan family, but it now seems quite 
evident that the resemblances are largely due to borrowing from 
Chinese. On the other hand, there seems to be some evidence for 
uniting the Kadai and Malayo-Polynesian families. 

24.26 The Malayo-Polynesian family is, after the Indo-Euro¬ 
pean, geographically the most widespread of all. It extends over 
most of the Pacific Islands and westward to Madagascar. Two 
branches may be recognized: Western or Indonesian includes most 
of the languages of the populous East Indies. Malay, originally 
the language of the north coast of Sumatra, has become widely 
extended throughout coastal Sumatra, Malaya, Borneo, and else¬ 
where. Indonesian, the official language of the new republic, is 
based largely on Malay, but with the admixture of elements from 
other closely related languages of the area. Javanese, Sundanese, and 
Maduran are the languages of Java. Of the many languages of 
Sumatra other than Malay, Balak is the best-known. Others in¬ 
clude Balinese on Bali, Dayak in the interior of Borneo, Makassar 
on Celebes. The whole Philippine area uses Indonesian languages: 
Tagalog, Bisayan, and Ilocano are the best-known. The Indonesian 
branch extends eastward to Guam ( Chamorro ) and westward to 
Madagascar (Malagasy). 

The Eastern branch is commonly divided into Micronesian, 
Polynesian, and Melanesian; however, these divisions are cer¬ 
tainly not coordinate. Polynesian is a closely related group of 
languages spoken over the expanse of the Pacific from Hawaii to 
New Zealand and Easter Island, as well as on a few islands much 
farther west. Hawaiian, Tahitian, Samoan, and Maori are the best 
known. Melanesian includes a large number of languages which 

364 Some Languages and Language Families 

are related, though commonly only distantly related. The best 
known is Fijian. 

24.27 Most of New Guinea and the interior of many adjacent 
islands is occupied by speakers of languages of unknown affinities. 
These are arbitrarily classified as Papuan. A genetic classification 
is greatly needed. 

24.28 The aborigines of Australia speak numerous languages 
which are related in a single family, Australian. The languages of 
Tasmania, now extinct, seem to have formed a separate family. 

24.29 The Dravidian languages are largely restricted to south¬ 
ern India. Four are spoken by very large populations and have 
highly developed literatures: Telegu, Tamil, Kannarese, and 
Malay dam. Brahui, spoken in Baluchistan at a considerable dis¬ 
tance from the main center of the family, figures prominently in 
discussion of the origin of the group. The hills of central India 
contain many tribes speaking Dravidian languages. The most im¬ 
portant are Gondi, Kurukh, and Kui. 

24.30 The Austro-Asiatic family consists of a number of lan¬ 
guages widely scattered through southeastern Asia and generally 
surrounded by languages of other families. In the hills of central 
India are found the Munda languages, the best known of which 
is Santdi. Khasi is spoken by a large tribe in Assam surrounded 
by speakers of Tibeto-Burman and Indie languages. Nicobarese is 
used on the Nicobar Islands in the Bay of Bengal. Pdauny and Wa 
occupy speech islands among the Tibeto-Burman and Kadai lan¬ 
guages of upper Burma. Mon is spoken in southern Burma where 
it is dominated by Burmese. Only Khmer and Vietnamese in Indo- 
China have any official standing. The latter is spoken by a very 
considerable population. 

24.31 The chief languages of America are Indo-European and 
have already been mentioned. In order of number of speakers they 
are English, Spanish, Portuguese, and French. One Indo-European 
language which originated in this hemisphere should be men¬ 
tioned. This is Haitian Creole, an offshoot of French, but suffi¬ 
ciently different that treatment as a mere dialect seems extreme. 
Besides these, other imported European languages, noticeably 
Italian and German, are spoken by more people in the Americas 
than any of the aboriginal American languages. Some of the latter 
do, however, show considerable vitality and continue to be used 


Some Languages and Language Families 

by large bodies of people. In South America four may be men¬ 
tioned : Guarani is the home language of most of Paraguay and of 
many in southwestern Brazil, though most of these are bilingual, 
using Spanish or Portuguese in public affairs. Quechua, the old 
language of the Inca Empire, is still used by several millions in 
Peru, Ecuador, and Bolivia, by no means all of whom can 
also speak Spanish. Aytnard is the language of many Indians 
in southern Peru and Bolivia. In Amazonian Brazil a lingua 
franca based on Tupi-Guarani and called Lingua Geral is now 
being replaced by Portuguese, but has been widely used in the 

In Middle America there are several Indian languages with 
appreciable numbers of speakers, though Spanish is making in¬ 
roads in most areas. In Mexico and Guatemala the following lan¬ 
guages are used by a hundred thousand or more: Nahuatl (nearly a 
million), Quiche, Cakchiquel, Mam, Yucaiec, Kekchi, Otomi, Zapolec, 
Mixtec, Totonac. North of Mexico Navaho is the only language even 
approaching such a figure. Of the smaller languages, some are 
rapidly losing ground and are faced with immanent extinction, a 
fate that has already overtaken a considerable number of lan¬ 
guages in the last three centuries. Some of the smaller speech com¬ 
munities are, however, holding their own. Occasional ones show 
considerable vitality. A few, like Navaho, have increased mark¬ 
edly in number of speakers. 

24.32 Listings of American languages are generally based on 
the situation at first European contact or earliest record. This is 
by no means contemporaneous over the whole hemisphere, but 
represents a convenient fiction. In many instances there have been 
great changes since. Not only have languages disappeared, but 
because of displacement of populations, many are now spoken in 
areas remote from their earlier location. The map on page 366 
and descriptions are based largely on the distribution at first 

A number of language families can be listed for North America, 
the exact number depending on the author’s interpretation where 
relationships are subject to debate. The classification adopted 
here is somewhat middle-of-the-road, by no means as conservative 
as some, but far from as radical as others. 

24.33 The eastern coast from Carolina northward to Labrador 

366 Some Languages and Language Families 

Some Language Groups in North America 

Some Languages and Language Families 367 

was occupied by speakers of Algonquian languages. These were, 
accordingly, the languages with which English and French had 
their first contacts. A considerable proportion of the Amerind loan 
words in these two languages are therefore from this source. A 
large number of place names are of Algonquian origin, even out¬ 
side the area originally occupied by the languages. For example, 
the Mississippi was explored from the north, and so bears an 
Algonquian name: missi Marge’ plus sipiy ‘water.’ 

Massachusetts, though a relatively minor language, is famous 
for Eliot’s Bible, the first modern missionary translation, and 
the first Bible printed in the United States. Powhatan, Delaware, 
Mohegan, Penobscot, Pasamaquoddy, and Micmac were other 
coastal tribes and languages. These were cut off from the main 
body of Algonquian languages by the Iroquois on the middle 
St. Lawrence. But to the north and west of this area Algonquian 
languages stretch to the Rocky Mountains. Four of these lan¬ 
guages, Fox (Wisconsin), Cree (Hudson Bay and westward), 
Menomini (upper Michigan), and Ojibwa (north shore of the Great 
Lakes) were used by Leonard Bloomfield in one of the first demon¬ 
strations of the applicability of comparative methods to languages 
without literary history. (See 23.13.) Potawatomi (lower Michigan), 
Illinois and Shawnee (Tennessee) are related. Blackfoot, Arapaho, 
and Cheyenne, spoken on the western plains, represent more dis¬ 
tantly related branches of the same family. 

24.34 Southeastern United States was an area of somewhat 
more diversity. The largest family was the Natchez-Muskogean: 
Creek, Alabama (including Koasati as a dialect), Chickasaw, 
Choctaw, Natchez. 

In the mountains was Cherokee, still one of the more important 
Amerind languages of the United States. Cherokee is widely known 
for its syllabary (see 21.11). This and Tuscarora on the Carolina 
coast were the southern outliers of the Iroquoian family, the main 
body of which occupied the middle St. Lawrence Valley and 
eastern Pennsylvania: Huron; Erie; Oneida; Mohawk; one language 
spoken by Seneca, Onondaga, and Cayuga; and Conestoga or Sus¬ 

In the Southeast also were Biloxi, Ofo, Tuielo and Catawba, 
outliers of the Siouan stock. This family occupied the greater part 
of the northern Great Plains: Dakota, Mandan, Winnebago, Chiwere 

368 Some Languages and Language Families 

(with dialects Iowa and Missouri), Dhegiha (with dialects Omaha, 
Ponca, Osage, Kansa, Quapaw or Arkansa), Hidalsa, Crow. 

To the south of the Siouan area, the main family was Caddoan, 
with Caddo, Wichita, and Pawnee as important languages. In the 
lower Mississippi valley, Tunica, Atakapa, and Chilimacha formed 
the Tunican family. In Tennessee there was Yuchi, forming a 
family of its own. 

The Natchez-Muskogcan, Iroquoian, Caddoan, Tunican, and 
Yuchian families are perhaps related, but proof has not yet been 
advanced. The Hokan languages of the West are usually included, 
and the proposed phylum is called Hokan-Siouan. 

24.36 The Arctic and Sub-Arctic coasts from Labrador and 
Greenland to Alaska are occupied by speakers of the Eskimo-Aleut 
languages. Aleut is spoken on the Aleutian Islands. In Alaska there 
is Inupik and from Alaska to Greenland, Yupik. These two are 
commonly called Eskimo. 

24.36 The Pacific Northwest from the Columbia River Valley 
to southern Alaska is occupied by a number of languages which 
share certain superficial peculiarities, noticeably very complex 
phonology. Many of them have very complicated patterns of 
clustering (see 16.14). The area is notable for the large proportion 
of the pioneer work in American linguistics which was done here. 
Franz Boas worked on a number of languages of the region, and 
particularly on Kwakiutl. Many of his co-workers and students 
worked on others. Most of the languages fall into three families 
which are related in one phylum known as Mosan. These are 
Salishan with Bella Coola, Coeur d'Alene, Chchalia, Kalispel; 
Wakashan with Nootka, Kwakiull, Bella Bella; and Chimakuan 
with Chimalcum and Quileule. There are four languages each of 
which forms a family of its own: Haida and Tlingit in Alaska, 
Tsimshian immediately to the south, and Kulenai on the eastern 
edge of the area. Each of these is sometimes assigned to a phylum 
containing other stocks (see 24.37 and 24.42). One of these some¬ 
what problematical phyla consists of Algonquian, Mosan, Kutenai, 
and Ritwan. The latter consists of two languages, Yurok and 
Wiyot in northern California. 

24.37 A large part of Oregon and California is a region of 
extreme linguistic diversity. In one recent classification 25 families, 
nearly one third of the whole number for the continent, were listed 

Some Languages and Language Families 369 

as restricted to this area. Except for Ritwan, these seem to be 
related in two phyla, Penutian and Hokan. The languages of 
proved Penutian affinity are all in this area, though there is some 
basis for believing that Tsimskian of British Columbia may be 
shown to be Penutian. Other, even less certain, affiliations have 
been proposed between Penutian and various languages or families 
in Middle America. The following languages, each representing a 
separate family, may be listed: Wintun, Maidu, Miwok, Costanoan, 
YokutSy Chinook, Kalapuya, Takelma, Siuslaxv, Coos. Chinook 
provided the base for Chinook Jargon, a creolized trade language 
that was once widely used throughout the polyglot Pacific North¬ 

24.38 The Hokan phylum includes a number of California 
languages, which, like the Penutian, are not very closely related, 
so that they are often classified in a number of families: Karok, 
Shasta, Chimariko, Yana, Pomo, Esselen, Salinan, Chumash. Yana 
is widely cited because of a paper of Edward Sapir on the differen¬ 
tiation between men’s and women’s speech. In Arizona and Lower 

( California is the Yuman group, containing a considerable number 
of languages. The Hokan languages extend far southward. 77a- 
pancc and Subtiaba are dialects of a single language spoken in 
southern Mexico and Nicaragua respectively. Tequistlatec or 
ChontaL of Oaxaca is also spoken in Southern Mexico. Jicaque is 
spoken in Honduras. The languages of northeastern Mexico are 
assumed to have been either Hokan, or to belong to a Coahuiltecan 
group allied to Hokan. The best preserved, Comecrudo, seems to be 
clearly Hokan, but on most there is so little data that no final 
answer is possible, 

Tonkawa is the only one of several languages of eastern Texas 
of which we have reliable record. It may be distantly related to 
Hokan. As we mentioned before, several families of central and 
eastern United States are sometimes joined with the Hokan to 
form the Hokan-Siouan phylum, but this must be considered as 
unproven and probably unprovable until a great deal more work 
has been done on the individual families. 

24.39 In Central America the most clearly marked and easily 
the most important language family is the Mayan. This occupies a 
compact area from the Isthmus of Mexico into Honduras. In the 
highlands of Guatemala are found Mam, Kekchi, Quiche, Cak- 

370 Some Languages and Language Families 

chiquel, Pokomam, Pokonchi, Ixil, and several lesser, languages 
forming a sub-family. To the west in Mexico are Tzellal, Tzotzil, 
Tojolabal, Choi, and Chontal of Tabasco. With this group belongs 
Chorti in Honduras, though it is the most eastern of the Mayan 
languages. To the north is YvcaUc. This area was the center of the 
pre-conquest Mayan empire, and it is assumed that the language 
of the empire was close to modern Yucatec. The only Mayan 
language not in the main area is Huaxtec, spoken on the Gulf coast 
of central Mexico. 

24.40 West of the Mayan area in southern Mexico is another 
region of extreme linguistic complexity. Most of these languages 
have been very poorly known until recently. The Summer Insti¬ 
tute of Linguistics now has active fieldwork underway in most of 
the languages and dialects, and accurate grammars and word lists 
are appearing. Comparative work is also getting started, and 
several small groups of proven relationship have been established. 
Previously, classification was usually based on general impressions 
of superficial similarities. 

Mixe, Zoque, and Popoluca of Vera Cruz form one such group. 
To the north are Totonac in several dialects and Tepehua forming 
another. These two with Huave are probably related to Mayan, 
and the larger group is sometimes referred to as the Macro-Mayan 

Zapotec, actually a group of related languages rather than a 
single language, and Chatino form another group where compara¬ 
tive reconstruction is clearly possible, though not yet published. 

Mixtec, Cuicatec, Trique, and Amusgo form another group in 
which comparative reconstruction is underway. Trique is becom¬ 
ing well known as the first language to be discovered with a tone 
system based on five levels. 

North of the last group is another composed of MazaUc, Chocho, 
Ixcatcc, and Popoloca of Pueblo. (Popoloca and Chontal are 
Nahuatl words meaning ‘barbarians’ and ‘foreigners.’ They are 
applied to many different languages, so that it is always necessary 
to add some further designation.) 

Still farther north are Olomi and the related Mazahua and Pame. 
To the west is Tarascan. 

Most of these groups have at one time or another been asserted 
to be related. Otomanguean is a name commonly applied, but 

Some Languages and Language Families 371 

differs somewhat in scope from author to author. There is some 
question as to how much of this collection of languages can ul¬ 
timately be related to one another. It does seem likely that 
Otomian, Mazatecan, and Mixtecan are related, and that the rela¬ 
tionship may soon be demonstrated. The Zapotecan and Tarascan 
seem more distant, if related at all. 

24.41 The most recent arrival in the area is Nahuatl, the lan¬ 
guage of the Aztec empire at the time of the conquest. The Nahuas 
established colonies throughout Middle America, at least as far 
as Nicaragua. Their language is still spoken in many areas scat¬ 
tered throughout the region. Nahuatl was one of the few Amerind 
languages to develop a writing system before European contact. 
Most of these documents have been lost, but there are many 
writings from the period immediately after the conquest in what 
is known as Classical Nahuatl. This language is the largest single 
source of the Amerind loan words in European languages. 

Nahuatl is the most southerly member of the Uto-Aztecan 
family. The center seems to be in the Great Basin and the Colorado 
Valley: Shoshone, Paiute, Tubatulabal, and Hopi. From southern 
Arizona into northwestern Mexico is a second large area with 
Papago, Pima, Tarahumara, Cora, and Huichol. Comanche was 
spoken in the southern Great Plains. 

24.42 The main area of the Uto-Aztecan family is interrupted 
by languages of the Athabaskan family. These are evidently an 
intrusion from the north; the main body of the family is in the 
far northwest of Canada and central Alaska. Here the best known 
are Sarsi and Chipewyan. That this is the center of diffusion is 
indicated by the fact that Haida and Tlingil, apparently related 
to the Athabaskan family, with which they form the Na-Den6 
phylum, are immediately adjacent. Moreover, the southern lan¬ 
guages form a compact group of closely related languages known 
collectively as Apachean. The most important of these is Navaho, 
the most extensively studied of all American Indian languages. 
There is also a small group of Athabaskan languages in northern 
California: Hupa, Chasta Costa, and Mattole. 

24.43 The southern Athabaskan languages are spoken by 
nomadic peoples surrounding the sedentary pueblo peoples. The 
latter speak a variety of languages, many of them restricted to a 
single village. These fall into three families: Tanoan, Keresan, and 

372 Some Languages and Language Families 

Zuni. The first certainly, and the last possibly, are related to the 
Uto-Aztecan family and form the Aztec-Tanoan phylum. 

24.44 Just as the smaller groups are frequently related to form 
larger groups, we may expect that some of the largest groups are 
related to each other at a still deeper level. Many suggestions have 
been made, none of them supported by conclusive evidence. For 
example, the Finno-Ugric and Altaic families are commonly united 
into a Ural-Altaic family, and this is sometimes extended to in¬ 
clude Japanese, Korean, and even Eskimo. Others have seen the 
relationships of Finno-Ugric with Indo-European, or with Dra- 
vidian, or with Austro-Asiatic. Another favorite hypothesis has 
been the relationship of Indo-European to Semitic. Of all such 
proposals, the best that can be said is that none are as yet proven. 
By the nature of the case it is impossible to disprove any such 
theory. We may expect that continued work will result in more 
comprehensive categories. However, in many of these proposals, 
the relationships postulated are at such a deep level that they are 
probably not demonstrable by direct comparison of present-day 
languages. Instead, they must await more thorough knowledge 
of the reconstructed ancestral forms of the various language 
families. It may be that when we have a clear picture of the older 
stages of some of the families we will be able to achieve important 
results by comparisons of these reconstructed languages. In the 
meantime, a cautious attitude must be maintained with regard 
to broader groupings of language families. 

Selected Bibliography 

American Council of Learned Societies, Committee on the Language 
Program. Structural Notes and Corpus. 1952. 109 pp. 

Atwood, Elmer Bagby. A Survey of Verb Forms in the Eastern United 
States. Ann Arbor: University of Michigan Press, 1953. 53 pp. plus 

Bloch, Bernard, and George L. Trager. Outline of Linguistic Analysis. 

Baltimore: Linguistic Society of America, 1942. 82 pp. 

Bloomfield, Leonard. Language. New York: Holt, 1933. 564 pp. 

-. Outline Guide for the Practical Study of a Foreign Language. 

Baltimore: Linguistic Society of America, 1942. 16 pp. 

Boas, Franz. Race , Language, and Culture. New York: Macmillan, 1940. 
pp. 119-242. 

Carroll, John B. The Study of Language; A Survey of Linguistics and 
Related Disciplines in America. Cambridge, Mass.: Harvard Uni¬ 
versity Press, 1953. 2S9 pp. 

Cummings, Thomas Fulton. How to Learn a Language. New York, 1916. 
Fries, Charles Carpenter. American English Grammar. New York: 
Appleton-Century, 1940. 313 pp. 

-. Teaching and Learning English as a Foreign Language. Ann Arbor: 

University of Michigan Press, 1945. 153 pp. 

-. The Structure of English; An Introduction to the Construction of 

English Sentences. New York: Harcourt, Brace, 1952. 304 pp. 

Gelb, Ignace J. A Study of Writing; The Foundations of Grammatology. 

Chicago: University of Chicago Press, 1952. 295 pp. 

Gray, Louis Herbert. Foundations of Language. New York: Macmillan, 

1939. 530 pp. 

Hall, Robert A. Leave Your Language Alone! Ithaca, New York: 
Linguistica, 1950. 254 pp. 

Heffner, Roe-Merrill Secrist. General Phonetics. Madison: University of 
Wisconsin Press, 1949. 253 pp. 

International Institute of African Languages and Cultures. Practical 
Orthography of African Languages, rev. ed. London: Oxford University 
Press, 1930. 24 pp. 

International Phonetic Association. The Principles of the International 
Phonetic Association, Being a Description of the International Phonetic 
Alphabet and the Manner of Using It. London, 1949. 53 pp. 

Jones, Daniel. An English Pronouncing Dictionary (on Strictly Phonetic 
Principles). London: J. M. Dent, 1917. 419 pp. 

-. An Outline of English Phonetics, 6th ed. New York: E. P. Dutton, 

1940. 326 pp. 


374 Selected Bibliography 

Jones, Daniel. The Phoneme; Its Nature and Use. Cambridge: Heffner, 
1950. 267 pp. 

-. The Pronunciation of English, 3rd ed. Cambridge: Cambridge 

University Press, 1950. 206 pp. 

Joos, Martin. Acoustic Phonetics. (Language Monographs, No. 23). 

Baltimore: Linguistic Society of America, 1948. 136 pp. 

Kenyon, John Samuel. American Pronunciation; A Textbook of Phonetics 
for Students of English, 8th ed. Ann Arbor: George Wahr, 1940. 248 pp. 

-and Thomas Albert Knott. A Pronouncing Dictionary of American 

English. Springfield, Mass.: G. and C. Merriam, 1944. 484 pp. 
Kroeber, Alfred Louis. Anthropology; Race, Language, Culture, Psy¬ 
chology, Prehistory. New York: Harcourt, Brace, 1948. pp. 206- 

-. Anthropology Today. Chicago: University of Chicago Press, 1953. 

966 pp. 

Kurath, Hans. Handbook of the Linguistic Geography of New England. 
Providence: Brown University Press, 1939. 23S pp. 

-. A Word Geography of the Eastern United States. Ann Arbor: 

University of Michigan Press, 1949. 88 pp. plus maps. 

Meillet, Antoine and Marcel Cohen. Les Langues du Monde, rev. ed. 

Paris: Champion, 1952. 1296 pp. plus maps. 

Miller, George Armitage. Language and Communication. New York: 
McGraw-Hill, 1951. 298 pp. 

Nida, Eugene Albert. Bible Translating; An Analysis of Principles and 
Procedures with Special Reference to Aboriginal Languages. New York: 
American Bible Society, 1947. 362 pp. 

-. Morphology, The Descriptive Analysis of Words, 2nd ed. Ann 

Arbor: University of Michigan Press, 1949, 342 pp. 

-. Learning a Foreign Language; A Handbook for Missionaries. 

New York: Foreign Missions Conference of North America, 1950. 
237 pp. 

-. God’s Word in Man’s Language. New York: Harper, 1952. 191 pp. 

North, Eric McCoy. The Book of a Thousand Tongues. New York: 
Harper, 1938. 3S6 pp. 

Pike, Kenneth Lee. Phonetics, A Critical Analysis of Phonetic Theory 
and a Technique for the Practical Description of Sounds. Ann Arbor: 
University of Michigan Press, 1943. 182 pp. 

-. The Intonation of American English. Ann Arbor: University of 

Michigan Press, 1945. 200 pp. 

-. Phonemics; A Technique for Reducing Languages to Writing. 

Ann Arbor: University of Michigan Press, 1947. 254 pp. 

-. Tone Languages. Ann Arbor: University of Michigan Press, 1948. 

187 pp. 

Potter, Ralph K., George A. Kopp, and Harriet C. Green. Visible 
Speech. New York: Van Nostrand, 1947. 441 pp. 

Sapir, Edward. Language; An Introduction to the Study of Speech. 
New York: Harcourt, Brace, 1921. 258 pp. 

Selected Bibliography 375 

-. Selected Writings in Language, Culture, and Personality. Berkeley: 

University of California Press, 1949. 617 pp. 

Stetson, Raymond Herbert. Motor Phonetics, A Study in Speech Move¬ 
ments in Action. Amsterdam: North-Holland Publishing Company, 
1951. 212 pp. 

Sturtevant, Edgar Howard. An Introduction to Linguistic Science. 

New Haven: Yale University Press, 1947. 173 pp. 

Thomas, Charles Kenneth. An Introduction to the Phonetics of American 
English. New York: Ronald Press, 1947. 181 pp. 

Ward, Ida C. Practical Suggestions for the Learning of an African Lan¬ 
guage in the Field. London: Oxford University Press, 1937. 39 pp. 

-. The Phonetics of English. 4th ed. Cambridge: Heffer and Sons, 

1948. 255 pp. 

Weinreich, Uriel. Languages in Contact: Findings, and Problems. New 
York: Linguistic Circle of New York, 1953. 148 pp. 

Welmers, William E. Spoken English as a Foreign Language. Wash¬ 
ington, D.C.: American Council of Learned Societies, 1953. 27 pp. 
Westermann, Dietrich, and Ida C. Ward. Practical Phonetics for Stu¬ 
dents of African Languages. London: Oxford University Press, 1933. 
227 pp. 

Whorf, Benjamin Lee. Four Articles on Metalinguistic)s. Washington: 
Foreign Service Institute, Department of State, 1950. 45 pp. 

Popular and Semi-Popular Introductions: A large number of intro¬ 
ductions for general readers have been published. Most of these are 
either worthless or worse. Sapir’s Language, though old, is still excellent, 
and is quite readable. Hall’s Leave Your Language Alone! represents the 
orthodox position of American linguistics in its strong reaction to 
traditional notions. 

General Treatments: Bloomfield's Language is the standard work. 
It covers most aspects of the field rather well, though it is somewhat 
contemptuous of phonetics. There are several peculiarities in which later 
authors have not followed Bloomfield, particularly his concern with 
psychologic theory. There is a good introduction to the comparative and 
historical method. Gray’s Foundations of Language is devoted primarily 
to this approach and is the best summary treatment of older historical 
methods. Sturtevant’s Introduction is precisely that — an introduction. 
It is readable and up-to-date and can be highly recommended to stu¬ 
dents. Sturtevant gives an interesting treatment of lapses, a subject 
neglected by others. Sapir’s Language is an older work still valuable 
because of the importance of the author and some special subjects 
treated. Nida’s Learning a Foreign Language, in spite of the restricted 
purpose expressed in its title, is an excellent comprehensive introduction. 
Carroll’s The Study of Language is primarily a statement of the organiza¬ 
tion of linguistics as a discipline and its place beside psychology, so¬ 
ciology, philosophy, and anthropology. Incidental to this main purpose, 
it gives a succinct statement of modern linguistic theory. 

376 Selected Bibliography 

Wider Point of View: Many anthropology texts contain brief dis¬ 
cussions of language from a cultural point of view. Krocber’s An¬ 
thropology is probably the best of these. Boas and Sapir were both 
leading anthropologists and linguists; some of their papers, reflecting 
their wide interest and contributing importantly to the development of 
American linguistics, arc reprinted in Race, Language and Culture and 
Selected Writings. Whorf’s Four Articles on Metalinguistics is a reprint 
of writings which have been very suggestive on the question of inter¬ 
relationship between language and other aspects of culture. Whorf’s 
hypothesis is currently the subject of intense debate among linguists 
and anthropologists. Kurath’s Handbook shows the close connection be¬ 
tween dialect investigations and social history. Weinreich’s Languages 
in Contact gives a comprehensive treatment of competition between 
languages and the resulting linguistic and social adjustments. 

Descriptive Technique: Bloch and Trager's Outline is a very good, 
but ovcrcondensed, statement of the analytic method of descriptive 
linguistics, representing more or less the orthodox American viewpoint. 
The series of textbooks of the Summer Institute of Linguistics gives an 
excellent treatment of analytic method with more emphasis on practical 
applications. Pike’s Phonemics and Nida’s Morphology are the two basic 
texts designed for technical courses in analytic method. Pike’s Tone 
Languages (in the same series) is unique in its field. Nida’s Bible Trans¬ 
lating gives detailed treatment of a number of technical points, par¬ 
ticularly meanings. Jones’ The Phoneme states a point of view which I 
would consider quite inadequate, but which is commonly reflected in 
writings of Jones’ followers (who are numerous) and hence cannot be 

Other Branches of Linguistics: Besides the treatments in the general 
works (particularly Bloomfield’s Language, Sturtevant's Introduction 
and Gray’3 Foundations of Language ), the following special works on 
specific branches of linguistics or closely related disciplines should be 
noted. There is no good reference work on the classification and enumer¬ 
ation of languages in English. Meillet and Cohen’s Les Langues du 
Monde, somewhat revised from a 1924 edition, is the best available. The 
best statement of the methods and problems is that of Greenberg in 
Kroeber’s Anthropology Today. 

The two works of Kurath, Handbook of the Linguistic Geography of 
New England and Word Geography, together with Atwood’s A Survey of 
Verb Forms, provide the best introduction to dialect studies for Amer¬ 
ican students. 

Miller’s Language and Communication is the best introduction to the 
psychological study of language. 

Phonetics: The best theoretical treatment of phonetics is Pike’s 
Phonetics. This is rather technical for most purposes. A more useful 
statement from the same point of view will be found in the introduction 
to Pike’s Phonemics. Heffner’s General Phonetics is probably the most 
satisfactory general textbook in the field, but still far from what is 

Selected Bibliography 377 

needed in that the emphasis is too strongly on sounds employed in 
familiar European languages. Westermann and Ward's Practical Pho¬ 
netics is indispensable for anyone dealing with African languages, and 
represents the phonetic theory current among many British phoneti¬ 
cians. Kenyon’s American Pronunciation is a comprehensive and de¬ 
tailed textbook, based largely on Midwestern American, but also making 
reference to other dialects. Jones’ Outline does the same for southern 
British. Kurath’s Handbook describes the detailed recording used in 
dialect work. The Principles of the International Phonetics Association 
gives a full statement of the IPA and its use. Joos’ Acoustic Phonetics 
is the best available general statement in this field. As a pioneer effort 
in a rapidly moving field, it is already somewhat out of date, but as yet 
there is nothing to replace it. Potter, Kopp and Green’s Visible Speech 
is a less general work, profusely illustrated with excellent plates. The 
latter make it indispensable for a beginner, but the phonetic transcrip¬ 
tions in the labels are sometimes difficult to interpret. Stetson’s Motor 
Phonetics represents a quite different approach which is just beginning 
to influence linguistic thought. 

Writing Systems: Pike’s. Phonemics gives a detailed discussion of con¬ 
siderations important in setting up an orthography. Nida’s Bible 
Translating gives some briefer but also very valuable suggestions. The 
International Institute’s Practical Orthography gives detailed recom¬ 
mendations for African languages. North's Book of a Thousand Tongues 
gives facsimile reproductions of Bible translations with historical notes 
and it is a useful reference on orthographies in use. Gelb’s A Study of 
Writing is the best treatment of the history of the alphabet and of other 
writing systems. 

Language Learning: Nida’s Learning a Foreign Language, Ward’s 
Practical Suggestions, and Bloomfield’s Outline Guide are all generally 
useful and in about the order stated. All are inexpensive, and anyone 
facing language learning in the field will find it worthwhile to have 
and study all. Cummings’ How to Learn a Language is older and hard to 
get, but is still useful for detailed directions for drill, etc. It has been a 
very important book in promoting better language study among mis¬ 
sionaries for many years. 

Mission Work: Few books have been written dealing directly with the 
linguistic problems that arise in mission work. Nida’s Bible Translating 
is intended as a handbook for the technical task of translating the 
Scriptures, particularly into languages which have not had them. It is, 
however, more widely useful, since the problems discussed are not 
settled once the Bible has been printed but must be continually faced 
by those who would use the new translation and those who are attempt¬ 
ing to present the same message in other forms. Nida's God's Word in 
Man’s Language is an excellent popular statement of the problems. 

English Language: All general works on linguistics written in English 
inevitably discuss various features of the language, often very profitably. 
A number of treatments of English phonology were listed and discussed 

378 Selected Bibliography 

in Chapter 17. The fuller evaluations of that chapter need not be re¬ 
peated here. Smith and Trager’s Outline of English Structure is the most 
satisfactory statement of English phonology and morphology, but the 
material is so condensed as presented that it is quite difficult. Fries’ 
American English Grammar should be indispensable for anyone teaching 
English, as it discusses actual observed usages in carefully selected 
groups of Americans. Fries’ Structure of English is an attempt at a 
formal analysis of English syntax. Its major shortcoming is inadequate 
use of stress and intonational data, with a resulting necessity of too free 
appeal to meaning, and a complex notation. The American Council of 
Learned Societies, Structural Notes and Corpus, is an outline for a series 
of English textbooks in various languages. It is a useful reservoir of in¬ 
formation on English phonology and grammar but its special purpose 
makes necessary a rather inconvenient organization. Welmers' Spoken 
English as a Foreign Language is a guidebook for the teacher using these 
textbooks and contains many suggestions of value in any teaching 
situation. Fries’ Teaching and Learning English is the most comprehen¬ 
sive treatment in English on language teaching, and embodies generally 
sound method but a somewhat outmoded analysis of English. Kurath’s 
Word Geography and Atwood's A Survey of Verb Forms in the Eastern 
United Slates are interesting, but somewhat restricted, discussions of 
American dialects, and may be useful for students who desire deeper 
understanding of their own speech form. 

The most useful periodicals for American students are the following: 

Language [Linguistic Society of America] 

International Journal of American Linguistics 


Studies in Linguistics 

The Bible Translator 

American Speech 

In addition there are frequently important articles on linguistic 
subjects in anthropological and orientalist periodicals. 

Access to recent literature is best had through the following bibliog¬ 

Comity International Permanent de Linguistes. 

Bibliographic Linguistigue des Annies 1939-1947. 

Utrecht: Spectrum, 1949-1950, 2 vols. 

This is continued annually by a volume such as the following: 

Comity International Permanent de Linguistes. Bibliographic Lin- 

guistique de l’Annie 1948 et Compliment des Annies 1939-1947. 

Headings are in both French and English. The coverage is very wide. 


Examples used in the text are from the following sources, which are hereby 

Andrade, M. J. 1933. “Quileute,” HAIL 3:149-292. 

Armstrong, L. E. 1940. The Phonetic and Tonal Structure of Kikuyu. 

Atwood, E. B. 1953. A Survey of Verb Forme in the Eastern United States. 
Bell, C. A. 1919. Grammar of Colloquial Tibetan. 

Blake, F. R. 1925. A Grammar of the Tctgdlog Language. 

Bloomfield, L. 1946. “Algonquian,” VFPA 6:85-129. 

Crowell, E. E. 1949. “A Preliminary Report of Kiowa Structure,” IJAL 

dc Francis, J. 1950. Nationalism and Language Reform in China. 

Diringer, D. 1948. Alphabet: A Key to the History of Mankind. 

Forrest, R. A. D. 1948. The Chinese Language. 

Garvin, P. L. 1948. “Kutenai I: Phonemics,” IJAL 14:37-42. 

Gleason, H. A. 1935. “ Melastomace*,” in Pulle, A. Flora of Surinam. 

Hall, R. A. 1944. Hungarian Grammar. 

Hall, R. A. 1948. Descriptive Italian Grammar. 

Hockett, C. T. 1952. [Class notes for Comparative Algonquian]. 

Hodge, C. T. 1946. “Serbo-Croatian Phonemes,” Language 22:112-120. 
Liljeblad, S. 1950. “Bannock I: Phonemes,” IJAL 16:126-131. 

Martin, S. A. 1951. “Korean Phonemics,” Language 27:519-533. 

Newman, S. 1947. “Bella Coola I: Phonology,” IJAL 13:129-134. 

Reichard, G. A. 1933. "Cceur d’Alene,” HAIL 3:519-707. 

Rosey, M. [n.d.] French Irregular Verbs. 

Swanton, J. R. 1911. “Tlingit,” HAIL 1:159-204. 

Voegelin, C. F., and Ellinghausen, M. E. 1943. “Turkish Structure,” JAOS 

Welmers, W. E. 1952. “Notes on the Structure of Bariba,” language 28:82- 

Westermann, D., and Ward, I. C. 1933. Practical Phonetics for Students of 
African Languages. 

Also the following persons: Martha Gleason, J. Maurice Hohlfeld, Paul 
Leser, Obadiah Manjengwe, Samuel Mentee, Paul Nilson, Howard Olson, 
Dorothy Peck, Wesloy Sadler, Henry Lee Smith, Jr. 

6.4, p. 67. The Hebrew examples here and elsewhere represent Massoretic 
Hebrew, that is, the Old Testament consonantal text as interpreted by the 
Massoretic tradition embodied in the vowel points and other punctuation. 
The transcription is one designed for pedagogic purposes. It represents what 
may be a possible phonemic system underlying the pointing, which will serve 
as a basis for an exposition of the grammar and provide a pronunciation which 
can be used in class. It is not intended that this system should fit either 
Modern Hebrew or the pronunciation at the time the Old Testament was 


380 Notes 

written. Certain inconsistencies in the pointing are resolved by assuming that 
the features shown arc not all contemporaneous and that there are a few 
instances of arbitrary writing conventions. Eleven vowel nuclei are recognized: 
i ii c ee a aa o oo u uu o. Stresses are written ' '. 

6.19, p. 74. The suggested notation, / aw «- (ay) /, is basically that of Nida. 
However, I have added the parentheses to make it somewhat clearer that the 
/ ay / does not occur in the form containing / aw *— (ay) /. This has been 
a source of confusion to students. 

10.9, p. 132. The terminology here is not satisfactory. There is much to 
commend the other usage (constitute for what is here called "construction," 
and construction for patterns of combination of constituents into constitutes). 
The difficulty, however, is that I have found "constitute” and "constituent" 
confusingly similar and often the cause of misunderstanding. Not desiring to 
propose a new term (nor, indeed, having any idea of one that would be 
suitable), I use this older set of terms. Still, "constitute" does need replace¬ 
ment by one with higher redundancy. 

16.8, p. 227. The "centering diphthongs” of Jones and Ward arc omitted 
from consideration since it is not clear to me what the true nature of these 
may be. It seems evident that there is something more than vowels and semi¬ 
vowel glides involved, perhaps / + /, or even terminals. 

18.10, p 258. Sources for these quotations are notcited. There is no intention 
to condemn individual works, but only to illustrate common failings which 
may be found even in otherwise excellent books. The works from which these 
examples were abstracted vary from utterly hopeless to far above the average. 
At least one of them shows an appreciation of the problems of language learn¬ 
ing which is quite unusual. The failing in this instance, as in so many, is 
insufficient knowledge of English phonology. This underscores the point made 
in 18.5, p. 254. 

18.15, p. 261. It must be distinctly understood that a test such as this is not 
infallible. Languages do have irregularities, but very often they are caused by 
inadequate analysis. Unfortunately, sometimes the converse is true; an inade¬ 
quate understanding of phonemics may conceal some morphophonemic diffi¬ 

18.17, p. 263. The bell-shaped curves should be interpreted as merely con¬ 
ventional representations of scatter. I have no desire to maintain that the 
distributions are actually in detail as shown. 

19.3, p. 267. I adhere closely to Longfellow’s version, however free he may 
have been with history. 

19.4, p. 268. I am not unmindful that information theorists are more 
accustomed to speaking of the capacity of a channel. But it seems to me that, 
for linguistic purposes, the concept of the capacity of a code is much more 
useful and a better place at which to begin an exposition of the elementary 

24.19, p. 358. Joseph Greenberg, whom I follow throughout in classification 
of African Languages, has proposed uniting his Eastern Sudanic and Central 
Sudanic into a "Macro-Sudanic" phylum. As, however, this name would seem 
to imply a group comprehending Sudanic, it seems unsatisfactory. I propose 
the name "Chari-Nile" as a substitute on the analogy' of his " Niger-Congo." 
I also prefer "Nilotic” to his Eastern Sudanic. 


Abkhasian, 356 
Acoli, 358 

acoustic phonetics, 13, 205-220 

addition of phonemes, 86 

adequacy of language, 5,126-127,155 

adjectivals, 96, 136 

adjectives, 95, 104 

adverbials, 137 

affixes, 58 

affricates, 22, 196 

Afghan, 354 

Africa orthography, 315 

Afrikaans, 351 

Afro-Asiatic languages, 345, 356 

Akan, 359 

Akkadian, 357 

Alabama, 367 

Albanian, 354 

Aleut, 368 

Algonquian languages, 340-342, 367, 

allographs, 302 

allomorphs, 61, 78-91 

allophones, 164, 176, 255 

alphabetic writing, 306, 311-317 

Altaic languages, 356, 372 

alveolae, 190 

alveolar sounds, 23, 192 

alveopalatal sounds, 23, 192 

Amharic, 357 

Amoy, 361 

amplitude, 208 

Amusgo, 370 

analogic change, 290, 297 

analysis, 56, 65-91, 128-142,172-186 

Anatolian languages, 355 

anatomy, oral, 188, 190 

Anglo-Saxon, 351 

animate gender, 148 

Annamese, 309 

Apachean languages, 371 

apex, 191 

apical sounds, 192 

Arabic, 84, 161, 185, 357 

Aramaic, 357 

Arapaho, 367 

Arkansa, 368 
Armenian, 354 
articles, 154 
articulations, 23, 192 
articulations, double, 199 
articulators, 23, 190 
articulatory phonetics, 13, 20, 35-39, 
187-204, 244 
aspirates, 22, 195 
Assamese, 355 
assimilation, 83, 85, 177 
Assyrian, 357 
Atakapa, 368 
Athabuskan languages, 371 
Australian languages, 364 
Austro-Asiatic languages, 364, 372 
Avar, 356 
Avestan, 354 
Aymarii, 365 

Aztec-Tanoan languages, 372 
Azerbaijani, 356 

Babylonian, 357 
back of tongue, 191 
back vowels, 201-203 
Bagirmi, 358 
Balinese, 363 
Balochi, 354 
Baltic, 354 
Bambara, 359 
bands, 209 
Bannock, 184 
Bantu, 336, 344, 360 
Baoulc, 359 
Bariba, 149 
base forms, 82, 87 
Basque, 356 
Bassa, 4, 359 
Batak, 363 
Beja, 358 
Bella Bella, 368 
Bella Coola, 246, 368 
Bemba, 360 
Bengali, 355 
Berber, 358 
bilabial sounds, 23, 192 

382 Index 

Biloxi, 367 

binits, 268 

Bisayan, 363 

bits, 268 

Blackfoot, 367 

Bloch, Bernard, 223 

Bloomfield, Leonard, 223, 367 

Boas, Franz, 368 

Bodo, 361 

borrowing, 290 

boundaries, dialectal, 294-296, 336 

Brahui, 364 

breath-groups, 203 

Breton, 351 

Bulgarian, 354 

Bulora, 359 

bundles of isoglosses, 293, 336 
Bushman, 361 
Byelorussian, 352 

Caddo, 368 

Caddoan languages, 368 

Cakchiquel, 365, 369 

Cantonese, 363 

capacity of a code, 268-270 

capitalization, 303 

cardinal vowels, 201 

cases, 139 

Catalan, 351 

Catawba, 367 

categories, 143-157 

Caucasian languages, 356 

Cayuga, 367 

Celtic languages, 351 

Central Saharan languages, 359 

Central Sudanic languages, 358 

Chad languages, 358 

Chaga, 360 

Chamorro, 363 

change, 6, 287-291, 326 

channels, 267 

Chari-Nile languages, 346, 358 
Chasta Costa, 371 
Chatino, 370 
Chechen, 356 
Chchalia, 368 
('heremiss, 356 
Cherokee, 307, 367 
Cheyenne, 367 
Chickasaw, 367 
Ohimakuan languages, 36S 
Chimakum, 368 

Chimariko, 369 

Chinese, 7, 144, 241, 305-306, 308, 
316, 361 

Chinook, Chinook Jargon, 369 

Chipcwyan, 371 

Chitimacha, 368 

Chiwere, 367 

Chocho, 370 

Choctaw, 367 

Choi, 370 

Chontal of Oaxaca, 369 
Chontal of Tabasco, 370 
Chorti, 370 
Chumash, 369 

classes and classing, 58, 65, 78-91, 
92-95, 136-137 
Classical Nahuatl, 371 
classification, 333-349 
clauses, 46 

clause terminals, 46, 138 
clicks, 199 
Coahuiltecan, 369 
code, 267 

Cceur d’Alene, 2-16, 368 
cognates, 339 
color terms, 4 
Comanche, 371 
Comecrudo, 369 
common elements, 335, 338 
communication, 266-283 
comparative method, 11, 338-343 
complementary distribution, 80, 164 
compounds, 59, 109 • 

concord, 140-142, 152 
conditioning, 61 
Conestoga, 367 
consonant charts, 24, 200 
consonants, 14-26, 193-200, 223- 
226, 247, 249 
constituent classes, 136 
constituents, 132 
constructions, 132 

content, 3-5, 12, 54-55, 75, 143-157 

contiguous assimilation, 84 

contours, 48, 108, 138 

Coos, 369 

Coptic, 357 

Cora, 371 

corpus, 65 

correspondences, 296, 339 
Costanoan, 369 
count nouns, 145 

Index 383 

cps, 206 

Crce, 116-122,124-125,148,151-153, 
308, 340-342, 367 
Creek, 367 

Creole, Haitian, 352, 364 
Croatian, 352 
Crow, 368 
cues, 218 
Cuicatec, 370 
Cushitic languages, 358 
cycles per second, 206 
Czech, 352 

Dakota, 367 

Danish, 327, 351 

Dayak, 363 

decoder, decoding, 267 

Delaware, 367 

demonstratives, 153-154 

density of communication, 293 

dentals, 23, 192 

derivation, 96, 107 

descriptive linguistics, 11, 284-287 

Dhegiha, 368 

dialects, 27, 31, 62, 146, 228, 232, 
285, 291-300 
Dinka, 242, 358 
diphthongs, 29, 202 
discontinuous morphemes, 72 
dissimilation, 85 
distribution, 56 
dorsals, 192 
dorsum, 191 

double articulations, 199 
dramatic dialects, 299 
Dravidian languages, 364, 372 
duals, 147 
Dutch, 320, 351 

Efik, 360 
Egyptian, 357 
encoder, encoding, 267 
English, 350, 364. Only major refer¬ 
ences arc given here. Sec under 
content, 144-155 
dialects, 31-34, 294-297, 299 
list of phonemes, 50 
morphology, 92-110 
phonology, 14-50, 215-217, 221- 
237, 245-250 
syntax, 128-140 

Erie, 367 

Eskimo, 368, 372 

Eskimo-Aleut languages, 368 

Esselen, 369 

Estonian, 356 

Ethiopic, 357 

Etruscan, 352 

exclusive first persons, 151 

expression, 3-5 

Ewe, 359 

eye dialect*, 299 

fading terminals, 46 
Fanti, 359 

fascicles of isoglosses, 293 
Fijian, 364 
Finnish, 356 

Finno-Ugric languages, 356, 372 

fit, 302 

flaps, 199 

Flemish, 320, 351 

Fon, 359 

Foochow, 361 

formant, 212 

fortis sounds, 195 

Fox, 340-342, 367 

free variation, 163, 181,262-264 

French, 123-125, 241, 242, 351, 364 

frequency, 206 

friction, fricatives, 21, 189-194, 220 

Fries, C. C., 223 

Frisian, 350 

front of tongue, 191 

frontal sounds, 192 

Fukien "dialects," 361 

Fulani, 359 

functional load, 232 

fundamentals, 207 

Gaelic, 351 
Galician, 351 
Galla, 358 
Ganda, 360 
Garo, 361 
Ge’ez, 357 
gender, 147 
Georgian, 356 

German, 86, 156, 294, 318-320, 351, 

Germanic languages, 350 
glides, 33, 37-39, 183 
glottal sounds, 21, 23, 189 

384 Index 

glottalized sounds, 189, 196 

glottochronology, 343 

Gondi, 364 

Gothic, 351 

government, 139-140 

grammar, 11, 52, 57, 88, 125-126 

graphemes, 302 

Greek, 60, 73, 85, 86, 302, 354 

Greek alphabet, 312 

groove fricatives, 22, 194 

Guarani, 365 

Gujcrati, 355 

Gur languages, 359 

Haida, 368, 371 
Haitian Creole, 352, 364 
Hakka, 361 

Hamitic languages, 344-345, 356 

Hamito-Semitic languages, 356 

harmonica, 207 

Hatsa, 361 

Hausa, 358 

Hawaiian, 363 

heads, 141 

Hebrew, 59, 67-73, 80-81, 86, 89, 
90, 141-142, 147, 155, 314, 315, 

Herero, 360 
Hidatsa, 368 

Hindi, 161, 184, 241, 244, 254, 330, 

Hiragana, 310 
Hittite, 355 

Hokan languages, 368, 369 

Hokan-Siouan languages, 368, 369 

homonyms, 329 

homophones, 261, 316 

honorific*, 156 

llopi, 371 

Hottentot, 361 

Huave, 370 

Huaxtec, 370 

Huichol, 371 

Hungarian, 84, 356 

Hupa, 371 

Huron, 367 

Ibo, 323, 359 
Icelandic, 351 
ideograms, 304 
Illinois, 367 
Illyrian, 355 

Ilocano, 363 

immediate constituent, 133-140 
implosivcs, 197 
inanimate gender, 148 
inclusive first persons, 151 
Indie languages, 354 
Indo-European languages, 350-356 
Indo-Hittite languages, 356 
Indo-Iranian, 355 
Indonesian, 363 
Indonesian languages, 363 
infix, 73, 74 

inflection, 96, 97-106, 111-127, 143- 

information, 268 
intelligibility, mutual, 334 
interdental sounds, 192 
International Phonetic Alphabet, 202, 

intonation, 40-50, 138, 236-237, 329 
Inupik, 368 
Iowa, 368 

Iranian languages, 354 
Irish, 351 

Iroquoian languages, 367, 368 
irregular forms, 99-100, 124 
isoglosses, 291-294, 336 
Italian, 84, 177, 321-322, 351, 364 
Ixcatec, 370 
Ixil, 370 

Japanese, 308, 309-311, 361, 372 

Javanese, 363 

Jicaque, 369 

Jones, Daniel, 201, 223 

Kabardian, 356 
Kabyle, 358 
Kadai languages, 363 
Kalapuya, 369 
Kalispel, 368 
Kamba, 360 
Kanji, 311 
Kannarese, 364 
Kansa, 368 
Kanuri, 359 
Karen languages, 361 
Karok, 369 
Kashmiri, 355 
Katakana, 310 
Kazak, 356 
Kekchi, 365-369 

Kenyon, John S., 223 
Kercsan, 371 
Khasi, 364 
Khmer, 364 
Khoisan languages, 361 
Kikuyu, 184, 360 
Kirabundu, 360 
Kiowa, 239, 241 
Kirghiz, 356 
Koasati, 367 
Kongo, 360 

Korean, 241, 309, 361, 372 
Kpelle, 359 
Kru, 147, 359 
Kui, 364 

Kuki-Chin languages, 361 
Kurdish, 354 
Kurukh, 364 

Kutenai, 241, 243, 244, 368 
Kwa languages, 359 
Kwakiutl, 368 

labial sounds, 192 
labialization, 199 
labiodental sounds, 23, 192 
labiovelar sounds, 193 
Landsm&l, 351 

language, definition of, 334, 338 
Lao, Laoticn, 363 
Lappish, 356 
larynx, 188 

lateral sounds, 21, 194, 198 
Latin, 87, 125, 140-142, 149, 154, 

Latinxua, 316 
Latvian, 354 
lax vowels, 202 

learning, 7, 115, 122, 124, 159, 161, 
166, 233, 2-13, 251-265 
lenis sounds, 195 
levels of analysis, 175 
levels of speech, 299, 319 
levels of structure, 66 
Lingua Geral, 365 

linguistics, 2, 11, 61, 126, 139, 205, 
277, 333 
lips, 191 
Lithuanian, 354 
loan-words, 99-100, 247, 290 
logarithms, 269 

Loma, 122, 124, 165, 185, 239, 243, 
261, 359 

Index 385 

loss of phonemes, 86 
Luba, 390 
Luganda, 239, 360 

Macro-Mayan languages, 370 
Maduran, 363 
Maidu, 369 
Makassar, 363 
Malagasy, 363 
Malay, 126, 363 
Malayalam, 364 

Malayo-Polynesian languages, 346, 

Malinkc, 359 
Mam, 91, 243, 365, 369 
Manchu, 356 
Mandan, 367 
Mandarin, 361 
Mandingo languages, 359 
manner of articulation, 188 
Maori, 363 
Marathi, 354 
markers, 135, 155 
Masai, 358 
mass nouns, 145 
Massachusetts, 367 
Mattole, 371 
Mayan languages, 369 
Mazahua, 370 
Mazatec, 370 

meaning, 54, 55, 77, 79, 92-94, 109, 

median sounds, 21, 194, 198 
Melanesian languages, 363 
Mende, 359 

Mcnomini, 340-342, 367 
metathesis, 86 
Miao, 308, 363 
Micmac, 367 

Micronesian languages, 363 

Mimi, 347 

mimicry, 8 

Mingrelian, 356 

minimal pairs, 16, 18, 25, 181 

Minoan B, 308 

Missouri, 368 

Mi wok, 369 

Mixe, 370 

Mixtec, 365, 370 

Mohawk, 367 

Mohegan, 367 

Mon, 364 

386 Index 

Mongol languages, 356 
Mongolian, 356 
Mordvin, 356 

morpheme, 11, 51-127; definition 
of, 51, 61 

morphology, 58, 97 
morphophoncmic changes, 82, 261 , 

Moru, 358 

Mosan languages, 368 
Moasi, 359 
Munda, 364 

mutual intelligibility, 334 

NarDen$ languages, 371 
Naga languages, 361 
Nahuati, 365, 371 
Nandi, 358 
nasal sounds, 21, 197 
nasalized sounds, 198 
Natchez, 367 

Natchcz-Muskogean languages, 367, 

Navnho, 365, 371 
Nepali, 355 

New English Dictionary, 349 
Ngala, 360 
Nicobarese, 364 
Nida, Eugene A., 223 
Niger-Congo languages, 345, 359 
Nilotic languages, 358 
noise, 280 
nominals, 136 

non-contiguous assimilation, 84 
non-vocoids, 249 
Nootka, 368 
North Caucasian, 356 
Northern Sotho, 360 
Norwegian, 351 

notation, 17, 22, 34, 47, 63, 224, 

nouns, 95, 97-101 
Nuba, 358 
nucleus, syllabic, 28 
Nuer, 358 

number, 7, 144, 157 
Nupe, 359 
Nyamwesi, 360 
Nyanja, 360 

obviative person, 152 
octave, 206 

Ofo, 367 

Ojibwa, 241, 340-342, 367 

Old Church Slavonic, 354 

Old English, 351 

Old High German, 351 

Old Icelandic, 351 

Old Persian, 354 

Old Saxon, 351 

Omaha, 368 

Oneida, 367 

Onondagc, 367 

open transition, 42, 45, 249 

oral resonants, 198 

order of morphemes, 57 

order of phonemes, 219 

order of words, 135-140 

orders, 112 

Oriya, 355 

Osage, 368 

Oscan, 352 

Otomanguean languages, 370 
Otomi, 365, 370 
over-differentiation, 173, 260 
over-all pattern, 233, 336 

Pahlavi, 354 
Paiute, 371 
palatal sounds, 192 
palatalization, 199 
palate, 190 
Palaung, 364 
Pame, 370 
Panjabi, 355 
Papago, 371 

Papuan languages, 346, 364 
paradigms, 95 
paradigmatic classes, 95 
paradigmatic subclasses, 102 
parts of speech, 92-96, 136-137 
Pasamaquoddy, 367 
Pashto, 354 
Pawnee, 368 
Penobscot, 367 
Penutian languages, 369 
Persian, 354 
person, 150-153 
personal pronouns, 95 
Phoenician, 311, 357 
phoneme, definition, 16, 162, 168 
phonemes, 9, 14-50, 158-186, 218- 

phonemic change, 289-290 

Index 387 

phonemic transcription, 22 
phonetic change, 287-288 
phonetic transcription, 22 
phonetics, 13, 20, 35-39, 187-220, 

phonology, 11 
phrases, 137 

Phrygian, 355 

Pike, Kenneth L., 223 

Pima, 371 

pitches, 40-50, 236 

points of articulation, 23, 188, 192 

Pokomam, 370 

Pokonchi, 370 

Polish, 352 

Polynesian languages, 245, 363 
Pomo, 369 
Ponca, 368 

Popoloca of Pueblo, 370 
Popoloca of Vera Cruz, 370 
Portuguese, 351, 364 
Potawatomi, 367 
Powhatan, 367 
Prakrits, 355 
preaspiration, 197 
predicates, 137 
prefixes, 59 
prenasalization, 197 
prepalatal sounds, 192 
prepositional phrases, 137 
prepositionals, 137 
primary stress, 41 
principal parts, 88 
productive subclasses, 99, 124 
progressive assimilation, 83 
pronominals, 136 
pronouns, 95, 105 
Provencal, 352 

Proto-Central Algonquian, 340-342 
proximate person, 152 
punctuation, 324-325 
Punic, 357 

Quaker speech, 106 
Quapaw, 368 
Quechua, 365 
Quiche, 365, 369 
Quileute, 73, 157, 368 

received pronunciation, 227 
reconstructions, 340-342 
redundancy, 272-277 

reduplications, 90-91 
references, 303 
regressive assimilation, 84 
repetition, 8 
replacives, 74 
resonance, 215 
resonants, 21, 193, 197-198 
retroflex sounds, 192 
Revere, Paul, 267 
Rhaito-Romanic, 352 
Riksm&l, 351 
rising terminals, 46 
Ritwan, 368 
Romance languages, 351 
Romansch, 352 
roots, 58 
Roumanian, 351 
rules, 88 
Rundi, 360 
Russian, 352 
Rwanda, 360 

Salinan, 369 
Salishan languages, 368 
Samoan, 363 
Samoyed, 356 
sampling, 167 
Sandawe, 361 
Sango, 359 
Sanskrit, 355 
San tali, 364 
Sapir, Edward, 369 
Sardinian, 352 
Sarsi, 371 

Schriftdeutsch, 319 
Scots Gaelic, 351 
secondary articulations, 188 
secondary stress, 45 
segmentation, 65, 169, 183-186, 218, 

Semitic languages, 357, 372 
semivowels, 21, 29, 33, 37-39, 198, 

Seneca, 367 
sentences, 137 
Sequoya, 307 
Serbian, 352 
Serbo-Croatian, 246, 352 
Shan, 363 
Shasta, 369 
Shawnee, 367 
Shilh, 358 

388 Index 

Shilluk, 358 
Shona, 322, 360 
Shoshone, 371 
Siamese, 363 
sibilants, 194 
Sindhi, 242, 355 
Sinhalese, 355 
Sino-Tibetan languages, 361 
Siouan languages, 367 
Siuslaw, 369 

Slavonic, Old Church, 354 

slit fricatives, 22, 194 

Slovak, 352 

Smith, Henry L., 223 

social dialects, 298 

Somali, 358 

Songhai, 358 

Sotho, 360 

South Arabic, 357 

South Caucasian languages, 356 

Southern Sotho, 360 

Spanish, 178-181, 351, 364 

spectrograms, 209-218, 219-220 

spectrograph, 209 

speech and writing, 44, 138, 316 

spelling pronunciations, 331 

spelling reforms, 327 

stem formative?, 60 

stems, 59 

stops, 22, 193, 194-197, 217 
stress, 40-50, 108, 138, 236 
structure, 2, 186 
subjects, 137 
sub-minimal pairs, 181 
subsystems, 219, 247, 311 
Subtiaba, 369 
Suchow, 361 

Sudanic languages, 344-345 
suffixes, 59 
Sumerian, 357 
Sundanese, 363 
suppletion, 104 
suspicious pairs, 176 
Susquehanna, 367 
sustained terminals, 46 
Swahili, 185, 360 
Swazi, 360 

Swedish, 59, 327, 351 
syllabaries, 306 
syllabic writing, 306-311 
syllable nuclei, 28 
syllables, 204, 306 

symmetry, 168 
synonyms, 329 
syntactic classes, 95, 106 
syntax, 58, 97, 128-142, 326 
Syrian, 357 

Tagalog, 90, 363 
Tahitian, 363 
Tokelma, 369 
Tamil, 364 
Tanoan, 371 
Tarahumara, 371 
Tarascan, 370 
Tasmanian, 364 
teeth, 191 
Telegu, 364 
Temne, 359 
tense vowels, 202 
Tepehua, 370 
Tequistlatcc, 369 
tertiary stress, 42 
Thai, 363 

Thomas, Charles K., 223 

Tibetan, 156, 361 

Tibeto-Burman languages, 361 

TigrtS, 357 

Tigrifia, 357 

time-depth, 343 

Tiv, 360 

Tlapanec, 369 

Tlingit, 243, 368, 371 

Tocharian, 355 

Tojolabal, 370 

tone languages, 239 

tongue, 191 

Tonkawa, 369 

Totonac, 365, 370 

Tragcr, George L., 223 

transcription, 22, 186 

transition, open, 42, 45, 249 

triangle, vowel, 201 

trills, 199 

Triquc, 370 

Tsimshian, 368 

Tawana, 360 

Tuareg, 358 

Tubatulabal, 371 

Tungus, 356 

Tunica, 368 

Tunican languages, 368 

Tupi-Guaranf, 365 

Turkic languages, 356 

Index 389 

Turkish, 112-116, 124-126, 168, 
278-280, 282, 356 
TOrkomcn, 356 
Tuscarora, 367 
Tutelo, 367 
Twi, 359 
Tzcltal, 370 
Tzotzil, 370 

Ukrainian, 352 
Umbrian, 352 
Umbundu, 360 

under-differentiation, 173, 260-264 
union languages, 322-323 
Ural-Altaic languages, 372 
Urdu, 330, 354 
Uto-Aztccan languages, 371 
uvula, 191 
uvular sounds, 192 
Uzbeg, 356 

velar sounds, 23, 192 

velic closure, 190 

velum, 190 

Venda, 360 

Venetic, 352 

verbal auxiliaries, 104 

verbs, 95, 101-104 

Vietnamese, 364 

vocabulary, 6-7, 100, 251, 328 

vocal cords, 188-190 

vocoids, 249 

voice, 20 , 188 

Votyak, 356 

vowel harmony, 84, 116, 278-280, 

vowel triangle, 201 
vowels, 27-39, 201-203, 215, 217, 
226-236, 247-249 

Wa, 364 

Wakashan languages, 368 
Ward, Ida C., 223 
weak stress, 41 
Welsh, 351 

West Atlantic languages, 359 

whisper, 189 

white noise, 220 

Wichita, 368 

Winnebago, 367 

Wintun, 369 

Wiyot, 368 

Wolof, 359 

word order, 135, 140 

words, 43 

writing systems, 301-317 
written language, 10, 138, 301-322 
Wu "dialects,” 361 

Xhosa, 360 

Yana, 369 

Yao, 363 

Yiddish, 351 

Yokuts, 369 

Yoruba, 243, 359 

Yucatec, 365, 370 

Yuchi, Yuchian languages, 368 

Yuman languages, 369 

Yupik, 368 

Yurok, 36S 

Zande, 359 
Zapotec, 365, 370 
Zenaga, 358 
zero, 75 
Zoquc, 370 
Zulu, 243, 360 
Zuni, 372 


§' N '» Delhi, n 



Author— G1 e--\ son. H.A 

Title— Introduction to descriptiv 

Borrower No. 

Date of Issue Date of Return